Command reference

clor

Fast web research, AI text/image/audio inference, social-platform research, email, cloud files, encrypted secrets, and bulk domain name checking, and more

One CLI with many services: web research, inference across Claude/GPT/Gemini/ElevenLabs (text, images, audio), social-platform reads, email over IMAP/SMTP, cloud-stored files with Unix-style verbs, encrypted JSON secrets, and bulk domain name checking, and more. New here? Run `clor account login` first (browser approval; the issued API key is saved automatically and reused by every subcommand). Then run `clor <subcommand> --help` on any group to see flags, output shapes, and runnable examples. `clor --long-help` dumps every subcommand's help in one shot if you want the whole tree.

clor [flags]

clor account

Sign in, see who is signed in (whoami), credit balance, recharge settings, and per-service usage

Read-only views of the account behind the current API key: who is signed in, account name and recharge settings, credit balance, and recent per-service spend.

clor account [flags]

clor account claim

Complete a sign-in approval by polling the token printed by `account login`

Single poll by default: status=pending|approved|cancelled|expired. On approved, persist the issued API key and report who is signed in. With --wait, block-poll until a terminal state or TTL elapses.

clor account claim <TOKEN> [flags]

clor account credit

Inspect the account's prepaid credit balance

clor account credit [flags]

clor account credit show

Show the current credit balance in USD with the last-updated timestamp and dashboard URL

clor account credit show [flags]

clor account delegate

Grant, list, and revoke the members who can run commands as you with --impersonate

A delegate is a member allowed to run commands as you with the --impersonate flag, like sudo scoped to acting as you. Team admins can already act as any member, so delegating is for giving one specific member (a shared support or automation account) that power without making them an admin. Use when: - you want a support or automation account to run work as you - you want to see who can act as you, or who you can act as - you want to revoke a delegate you granted earlier Subcommands: add Allow a member to run commands as you list Show who can act as you and who you can act as remove Revoke a member's permission to run commands as you Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor account delegate [flags]

clor account delegate add

Allow a member to run commands as you

Pass the member's user id. A team admin can list member ids with admin user list. Once added, that member can run commands as you with the --impersonate flag. Re-adding an existing delegate is a no-op.

clor account delegate add <DELEGATE USER ID> [flags]

clor account delegate list

Show who can run commands as you (granted) and who you can run as (received)

Two sections. Granted is the members who may act as you. Received is the members you may act as, the user ids valid for --impersonate. JSON keys: granted, received.

clor account delegate list [flags]

clor account delegate remove

Revoke a member's permission to run commands as you

clor account delegate remove <DELEGATE USER ID> [flags]

clor account info

Show the account's name, id, auto-recharge state, and credits dashboard URL

clor account info [flags]

clor account key

Create narrower API keys scoped to specific services and an optional expiry

Issue a new API key that is no more powerful than the key you are signed in with. The new key's capabilities must be a subset of your own, and it can expire sooner than yours but never later. Use it to give a node or a single workflow only the access it needs. Use when: - you want a per-node credential that runs spaces but cannot edit them - you want a key that can reach only one service, like drive - you want a key that expires automatically Subcommands: create Issue a new scoped, optionally expiring key Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor account key [flags]

clor account key create

Issue a new API key scoped to the given capabilities with an optional expiry

Issue a new key whose capabilities are a subset of the signed-in key's. A capability is a service slug (drive) or a service and action (claw:run). Tags are the runner identity this key carries; a key needs a host= tag to claim and run claws on a node, plus any role tags. Tags are the creator's choice and need not fit the signed-in key's tags. The plaintext key is printed once and cannot be retrieved later, so capture it now.

clor account key create [flags]

clor account login

Sign in by approving a browser link, or pass a key directly with --key

Three modes: - --key clor_xxx: verify the supplied key and store it in the target profile (pair with --profile NAME to name the account). Scriptable, no browser. - No API key set: allocate an approval claim and print the URL plus the `clor account claim <TOKEN>` resume command. With --wait, block until approved and persist the issued key. - Key already set: verify it and report who is signed in. At an interactive terminal with --stdout-format text (default), the URL opens in the system browser and the command waits for approval. Pass --wait=false to print and exit; --no-browser (env CLOR_NO_BROWSER=1) to skip the browser pop.

clor account login [flags]

clor account logout

Sign out by removing the stored API key and node identity from this machine

clor account logout [flags]

clor account me

Show the user identity attached to the current API key (id, email, name, picture)

clor account me [flags]

clor account profile

List, switch, and remove the API-key profiles stored on this machine

Each profile is one API key for one account, the way AWS profiles work. The active profile is chosen by the --profile flag, then CLOR_PROFILE, then the persisted default profile. Add a profile with `clor account login --key clor_xxx --profile NAME`. Use when: - the user holds keys for several accounts and wants to see them - the user wants to change which account is used by default - the user wants to drop a profile they no longer use Subcommands: list List stored profiles with active and default markers use Set the default profile used when no flag or env is given remove Delete a stored profile Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor account profile [flags]

clor account profile list

List stored profiles with active and default markers

clor account profile list [flags]

clor account profile remove

Delete a stored profile and its key

clor account profile remove <NAME> [flags]

clor account profile use

Set the default profile used when no flag or environment override is given

clor account profile use <NAME> [flags]

clor account ssh-key

Add, list, and remove SSH public keys available when creating machines

Save SSH public keys once, then select their fingerprints whenever you create a machine. Each key is available only to your account in the active team. Use when - you want to save an SSH public key for repeated machine creation - you need the fingerprint for a machine creation request - you want to remove a key you no longer use Subcommands add Store an SSH public key for machine selection list List SSH keys available when creating machines remove Remove an SSH key by its fingerprint Output. Every subcommand supports --stdout-format text|jsonl|json. Text output is logfmt with an event leader.

clor account ssh-key [flags]

clor account ssh-key add

Store an SSH public key for machine selection

clor account ssh-key add <PUBLIC_KEY> [flags]

clor account ssh-key list

List SSH keys available when creating machines

clor account ssh-key list [flags]

clor account ssh-key remove

Remove an SSH key by its fingerprint

clor account ssh-key remove <FINGERPRINT> [flags]

clor account usage

Show recent usage with a per-service spend and request-count breakdown

Returns total cost in USD, request count, and one row per service (search, inference, drive, etc.) over a chosen window. Pick the window with --since.

clor account usage [flags]

clor agent

Save Claude and Codex settings, credentials, and skills, install them into your local agent

Everything about the coding agents your spaces run. Save reusable, typed settings for Claude or Codex and install them into your local agent, store the OAuth tokens and API keys a space authenticates with, and save SKILL.md skills your agents pick up in every space. Use when: - you want to save a Claude or Codex settings profile and reuse it - you want to push a saved settings profile into your local ~/.codex/config.toml or Claude settings files - a space must run as a specific Claude or Codex account - you want to save a SKILL.md once and have every space's agent use it Subcommands: config Save, edit, and install typed Claude and Codex settings credential Store coding-agent OAuth tokens and API keys skill Save reusable agent skills, install them into the local agents Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor agent [flags]

clor agent config

Save, edit, and install typed Claude and Codex settings on your team

Reusable, typed settings for a coding agent, saved on your team. Each config is one agent type (claude or codex) carrying a typed settings body. Save several per type, edit one atomically by exporting the full JSON, changing it, and importing it back, and install a config into your local agent by merging it into ~/.codex/config.toml or ~/.claude/settings.json and ~/.claude.json, plus the user-configurable behavior prompt into ~/.codex/AGENTS.md or ~/.claude/CLAUDE.md below Clor's managed space context. Use when: - you want to save a Claude or Codex settings profile and reuse it - you want to edit a saved config's model, reasoning effort, or sandbox mode - you want to push a saved config into your local agent Subcommands: create Save a new config of a given type list List the team's saved configs show Show one config by id or name export Print a config's full JSON for editing import Replace a config's body from edited JSON install Merge a config into the local agent's settings file delete Delete a config Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor agent config [flags]

clor agent config create

Save a new agent config of a given type

Creates one config of the given type under a name unique to your team. Supply the typed body as JSON on stdin or with --stdin-file; when omitted an empty template is saved that you can fill in later with export and import. Several configs per type are allowed.

clor agent config create [flags]

clor agent config delete

Delete an agent config by id or name

Permanently removes one saved config. This does not change native agent files written by an earlier installation.

clor agent config delete <ID> [flags]

clor agent config export

Print an agent config's full JSON body for editing

Prints the config's typed body as indented JSON with every field shown, including empty ones, so you can edit it and pipe the result back through `agent config import`. This is the export half of the atomic edit round trip; the body is the exact shape import reads.

clor agent config export <ID> [flags]

clor agent config import

Replace an agent config's body from edited JSON

Reads a JSON config body from stdin or --stdin-file and atomically replaces the config, the import half of the export, edit, import round trip. The whole body is swapped, never merged. The config type is immutable; pass --name to rename the config at the same time.

clor agent config import <ID> [flags]

clor agent config install

Merge an agent config into the local agent's native settings

Fetches the config from your team, checks the local agent binary is a supported minimum version or newer, then merges the config's managed keys into ~/.codex/config.toml or Claude's ~/.claude/settings.json and ~/.claude.json files. The agent's global instructions file, ~/.codex/AGENTS.md or ~/.claude/CLAUDE.md, gets an installer-managed block of Clor-owned space context followed by the config's user-configurable behavior prompt, both as plain text at the top, leaving every other byte in that file untouched. Outside a space only the behavior prompt is written and any earlier Clor context block is removed. It also trusts the space root and current directory in the agent's native project settings and completes Claude onboarding. Existing authentication, project, and unknown settings are preserved. Honors CODEX_HOME and CLAUDE_CONFIG_DIR.

clor agent config install <ID> [flags]

clor agent config list

List the team's saved agent configs by id, name, and type

Metadata for every config your team has saved: id, name, type, and schema version. Use `agent config show` or `agent config export` to see one config's typed body.

clor agent config list [flags]

clor agent config show

Show one agent config by id or name

Resolves the argument as a config id or, failing that, a config name, and prints its metadata and typed body. Use `agent config export` when you want only the body JSON to edit.

clor agent config show <ID> [flags]

clor agent credential

Store agent OAuth tokens and API keys for Claude and Codex, then run spaces as that account

Password manager for coding-agent credentials. Each credential is a sealed token (a Claude or Codex OAuth token, or an Anthropic or OpenAI API key) tagged with its runtime and stored under a name. Attach one to a space and the daemon injects the matching environment variable into the spawned agent at run time, so the agent authenticates as that account on any host without a local sign-in. The token stays encrypted in the vault and is never copied onto the space, so rotating it once updates every space that uses it. Credential types: Claude Subscription Token: --runtime claude --oauth-token (injects CLAUDE_CODE_OAUTH_TOKEN) Anthropic API Key: --runtime claude --api-key (injects ANTHROPIC_API_KEY) Codex Access Token: --runtime codex --oauth-token (injects CODEX_ACCESS_TOKEN) OpenAI API Key: --runtime codex --api-key (injects OPENAI_API_KEY) Use when: - a space must run as a specific Claude or Codex account, or on a host that is not signed in locally - you want to bring your own Claude/Codex subscription or Anthropic/OpenAI API key to an unattended space - you need to rotate a token in one place across every space that uses it Subcommands: create Save an agent OAuth token or API key under a name list List saved credentials with their runtime and type show Show one credential's runtime and type without revealing the token test Validate a saved credential's token against its provider update Rotate the token, type, or runtime of a saved credential delete Permanently remove a saved credential Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor agent credential [flags]

clor agent credential create

Save an agent OAuth token or API key under a name for spaces to authenticate with

Supply exactly one of --oauth-token or --api-key. The token form pairs with --runtime to pick the environment variable the daemon injects. claude with an oauth token sets CLAUDE_CODE_OAUTH_TOKEN, claude with an api key sets ANTHROPIC_API_KEY, codex with an oauth token sets CODEX_ACCESS_TOKEN, and codex with an api key sets OPENAI_API_KEY. Re-running with the same name rotates the credential in place. How to get each token: Claude Subscription Token (--runtime claude --oauth-token) Run Claude Code agents on your Claude subscription using an OAuth token 1. Install Claude Code and sign in to your Claude subscription 2. Run `claude setup-token` in your terminal to start the OAuth flow 3. Approve the request in the browser, then copy the token that begins with sk-ant-oat- Anthropic API Key (--runtime claude --api-key) Run Claude Code agents against the Anthropic API with a long-lived API key 1. Open the Anthropic Console API keys page https://console.anthropic.com/settings/keys 2. Create a new API key and copy the value that begins with sk-ant- Codex Access Token (--runtime codex --oauth-token) Run Codex agents with a ChatGPT access token 1. In the active Business workspace, open Workspace settings → Permissions & roles as an owner or admin https://chatgpt.com/admin/access-tokens 2. Under Access tokens, turn on Allow users to create access tokens 3. Under Codex Local, turn on Allow members to use Codex Local 4. Create an access token (a one-year expiration is recommended) and paste it here OpenAI API Key (--runtime codex --api-key) Run Codex agents against the OpenAI API with a long-lived API key 1. Open the OpenAI platform API keys page https://platform.openai.com/api-keys 2. Create a new secret key and copy the value that begins with sk-

clor agent credential create <NAME> [flags]

clor agent credential delete

Permanently remove a saved credential

Permanent: no soft-delete or recovery window. Spaces referencing this credential by name fall back to machine file credentials on their next run.

clor agent credential delete <NAME> [flags]

clor agent credential list

List saved agent credentials by name without revealing any token

Metadata only (id, name, runtime, type, last access); tokens stay encrypted in the vault. Runtime and type come from plaintext labels, so listing never decrypts a token and never updates a credential's last access. Use `credential show` to decrypt one credential.

clor agent credential list [flags]

clor agent credential show

Show one credential's runtime and type without revealing the token

Decrypts the credential to report its runtime and type and the environment variable the daemon would inject. The token value itself is never printed. Use `clor secret get` if you must read the raw token.

clor agent credential show <NAME> [flags]

clor agent credential test

Validate a saved credential's token against its provider

Decrypts the credential and checks its token. API keys are validated with a short live call to the provider; subscription tokens are checked for the expected format. The token value itself is never printed. The command succeeds whenever the check runs and reports the verdict in the valid field; valid=false means the token was rejected or could not be confirmed. A non-zero exit means the credential could not be read.

clor agent credential test <NAME> [flags]

clor agent credential update

Rotate the token, type, or runtime of a saved credential

Updates only the fields you pass. Supply a new --oauth-token or --api-key to rotate the token (at most one), and/or --runtime to retarget the agent. Unset fields keep their current value. The credential must already exist.

clor agent credential update <NAME> [flags]

clor agent skill

Save reusable agent skills on your team and install them into Claude Code and Codex

Reusable skills for your coding agents, saved on your team. A skill is one SKILL.md, inline markdown installed as <slug>/SKILL.md into each agent's skills directory, or a source URL stored for reference. Spaces attach your visible skills at create; inside a space, install writes them into ~/.claude/skills and ~/.codex/skills. Use when: - you want to save a SKILL.md so every space's agent picks it up - you want to sync a skills directory from a file tree (import) - a space environment should install its attached skills at startup Subcommands: create Save a new skill from a SKILL.md file or a source URL list List the team's saved skills show Show one skill by id, name, or slug export Print a skill's SKILL.md markdown import Upsert a skill from a SKILL.md file install Write a space's attached skills into the local agents uninstall Remove the skills a previous install wrote delete Delete a skill Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor agent skill [flags]

clor agent skill create

Save a new agent skill from a SKILL.md file or a source URL

Creates one skill under a name unique to your team. Pass a SKILL.md file to store its full contents as the skill markdown; the file's frontmatter fills the name and description unless the flags override them. Pass --url instead to store a source URL without a body.

clor agent skill create [FILE] [flags]

clor agent skill delete

Delete an agent skill

clor agent skill delete <ID> [flags]

clor agent skill export

Print an agent skill's SKILL.md markdown

Prints the skill's stored markdown verbatim, the editable form import reads back. A url skill has no markdown body; use `agent skill show` to see its source URL.

clor agent skill export <ID> [flags]

clor agent skill import

Upsert an agent skill from a SKILL.md file

Reads a SKILL.md file and saves it as a skill: created when absent, replaced whole when present. The file's frontmatter fills the name and description unless the flags override them. Matching is by --slug when given, otherwise by name. --slug is the official directory sync's key and is accepted only from the official team's own key.

clor agent skill import <FILE> [flags]

clor agent skill install

Write a space's attached skills into the local agents' skills directories

Fetches the skills attached to this space and writes each markdown skill as <slug>/SKILL.md under ~/.claude/skills and ~/.codex/skills (honoring CLAUDE_CONFIG_DIR and CODEX_HOME). The space id comes from CLOR_SPACE_ID or --space; authentication is the normal key resolution, which inside a managed space is the space identity key. Re-running converges: skills removed from the space are removed locally, tracked by a receipt so nothing installed another way is touched. URL skills are listed but not installed.

clor agent skill install [flags]

clor agent skill list

List the team's saved agent skills by id, name, and type

Metadata for every skill visible to you: your own, the team-shared rows, and the official directory. Use `agent skill show` or `agent skill export` to see one skill's markdown.

clor agent skill list [flags]

clor agent skill show

Show one agent skill by id, name, or slug

Resolves the argument as a skill id or, failing that, a skill name or official slug, and prints its metadata. Use `agent skill export` when you want only the SKILL.md markdown.

clor agent skill show <ID> [flags]

clor agent skill uninstall

Remove the skills a previous install wrote from the local agents

Removes every skill directory the install receipt lists under ~/.claude/skills and ~/.codex/skills, plus the receipt itself. Skills installed any other way are never touched. Purely local; no request is made.

clor agent skill uninstall [flags]

clor bluesky

Post to Bluesky and read it: timeline, feeds, threads, search, follow, like, notifications, direct messages

Speaks the AT Protocol directly to a Bluesky server, authenticated with a handle and an app password stored as a typed secret. Covers reading (timeline, custom and author feeds, threads, search, profiles), posting with rich-text links, hashtags, mentions, replies, quotes, images, and video, the social graph (follow, block, mute), engagement (like, repost), notifications, and direct messages. Use when: - the user wants to read their Bluesky timeline, a profile, a thread, or search posts - the user wants to post, reply, quote, like, repost, or delete - the user wants to follow, unfollow, block, or mute accounts - the user wants to read notifications or send and read direct messages Subcommands: account Manage saved Bluesky credentials timeline Read the home timeline feed List, read, and browse feeds thread Read a post and its replies post Create a post, reply, or quote pin Pin one of your posts to the top of your profile unpin Remove the pinned post from your profile card Preview a URL's link-card metadata before posting video Upload a video and track its transcode job delete Delete one of your posts search Search public posts profile Show, search, or update profiles follow Follow an account unfollow Stop following an account follower List who follows an account following List who an account follows block Block, unblock, and list blocked accounts mute Mute, unmute, and list muted accounts like Like a post unlike Remove your like repost Repost a post unrepost Remove your repost notification Read notifications and mark them seen chat Read and send direct messages Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky [flags]

clor bluesky account

Manage saved Bluesky credentials (handle, app password) used by every Bluesky subcommand

Credentials are stored as a typed secret (handle, app password, optional service host). Once saved, every Bluesky subcommand picks them up automatically (or via --account when more than one is configured). App passwords are generated in the Bluesky app under Settings, Privacy and Security, App Passwords. They bypass two-factor prompts, so no email-code step is needed. Never store the real account password here. Use when: - the user wants to register a Bluesky account for the CLI to use - the user wants to see which Bluesky accounts are saved - the user wants to inspect or rotate stored credentials - the user wants to remove a Bluesky account Subcommands: add Save or update a Bluesky account by handle and app password list List saved Bluesky accounts visible to the caller get Read one saved Bluesky account, with the raw JSON in json mode delete Permanently remove a saved Bluesky account by name Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky account [flags]

clor bluesky account add

Save or update a Bluesky account by handle and app password for use by other Bluesky subcommands

[NAME] defaults to the handle itself; re-running with the same name updates in place. The app password is generated in the Bluesky app under Settings, Privacy and Security, App Passwords. It looks like xxxx-xxxx-xxxx-xxxx, applies only to API access, can be revoked independently, and bypasses two-factor prompts. Pass it on --app-password directly, or pipe it via --stdin-format text to keep it out of shell history. --service only needs setting for self-hosted servers; the default entryway routes to whichever server actually holds the repository.

clor bluesky account add [NAME] [flags]

clor bluesky account delete

Permanently remove a saved Bluesky account by name

After deletion the other Bluesky subcommands can no longer use the account unless re-added or supplied via --credentials-file or --stdin-format json.

clor bluesky account delete <NAME> [flags]

clor bluesky account get

Read a saved Bluesky account; pass --stdout-format json for the raw JSON including the app password

Default logfmt drops the app password; `--stdout-format json` returns the raw bluesky-account JSON suitable for piping into `clor bluesky <SUBCOMMAND> --stdin-format json`.

clor bluesky account get <NAME> [flags]

clor bluesky account list

List saved Bluesky accounts with id and name

Metadata only (id, name, last access); app passwords stay encrypted server-side. Use `account get` to retrieve the full JSON.

clor bluesky account list [flags]

clor bluesky block

Block and unblock accounts, and list the accounts you block

Blocking hides both directions: the account cannot see or interact with your posts, and theirs disappear from your views. Use when: - the user wants to block or unblock an account - the user wants to see who they have blocked Subcommands: add Block an account by handle or DID remove Unblock an account by handle or DID list List the accounts you block Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky block [flags]

clor bluesky block add

Block one or more accounts by handle or DID in a single session

clor bluesky block add <HANDLE>... [flags]

clor bluesky block list

List the accounts you block

clor bluesky block list [flags]

clor bluesky block remove

Unblock one or more accounts by handle or DID in a single session

clor bluesky block remove <HANDLE>... [flags]

clor bluesky card

Preview a URL's link-card metadata before posting: title, description, thumbnail

Preview the link card Bluesky would build for a URL, without posting. The card service returns the title, description, and a resized thumbnail image URL, so you can see what a post's card will look like and decide whether to override a field or supply your own thumbnail. Use when: - the user wants to see a URL's card title, description, and thumbnail before posting - the user wants to check whether a URL has a thumbnail so they can generate one when it does not - a workflow builds a post's link card in steps rather than in one shot Subcommands: extract Extract the title, description, and thumbnail for a URL Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky card [flags]

clor bluesky card extract

Extract the title, description, and thumbnail Bluesky would show for a URL

Query the card service for the title, description, and thumbnail image URL Bluesky would show for <URL>. When image is empty the URL has no thumbnail; generate one and pass it to `clor bluesky post --link-card-thumbnail`. likely_type and error report how the card service classified the URL.

clor bluesky card extract <URL> [flags]

clor bluesky chat

Read and send direct messages: list conversations, read one, send a message

Direct messages are served by your own data server, not the public appview, so these subcommands proxy through it automatically. Use when: - the user wants to see their direct-message conversations - the user wants to read the messages in a conversation - the user wants to send someone a direct message Subcommands: list List your direct-message conversations read Read the messages in one conversation send Send a direct message to an account Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky chat [flags]

clor bluesky chat list

List your direct-message conversations with their members and unread counts

clor bluesky chat list [flags]

clor bluesky chat read

Read the messages in one conversation, oldest first

<CONVERSATION> is a conversation id from `clor bluesky chat list`. Messages are returned oldest first so a thread reads top to bottom.

clor bluesky chat read <CONVERSATION> [flags]

clor bluesky chat send

Send a direct message to an account by handle or DID

Opens or reuses the one-to-one conversation with <HANDLE> and sends <TEXT>. The recipient must allow direct messages from you.

clor bluesky chat send <HANDLE> <TEXT> [flags]

clor bluesky delete

Delete one of your own posts by its at:// URI

<URI> is an at:// post URI, e.g. the one printed by `clor bluesky post`. Only records in your own repository can be deleted.

clor bluesky delete <URI> [flags]

clor bluesky feed

List saved feeds, read a custom or pinned feed, and read an account's posts

Read the custom and pinned feeds you follow in the app, and the author feed of any single account. A custom feed is an algorithmic feed generator (the kind you pin in the Bluesky app); the author feed is one account's own posts and reposts. Use when: - the user wants to read one of their saved or pinned feeds by name - the user wants to read a feed generator by its at:// URI - the user wants to see which feeds an account has saved or pinned - the user wants one account's own posts and reposts Subcommands: list List your saved and pinned feeds read Read the posts in a custom or pinned feed author Read the posts authored by one account Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky feed [flags]

clor bluesky feed author

Read the posts authored by one account, newest first

<HANDLE> is a handle (you.bsky.social) or a DID. Returns that account's own posts and reposts, the same feed shown on their profile.

clor bluesky feed author <HANDLE> [flags]

clor bluesky feed list

List your saved and pinned feeds

Lists every feed saved to your account with a pinned= flag for the ones you have pinned. Feed-generator display names are resolved, so the name= attr matches what you see in the app. Pass a name or the uri= value to feed read.

clor bluesky feed list [flags]

clor bluesky feed read

Read the posts in a custom or pinned feed, newest first

<FEED> is an at:// feed-generator URI, or the display name of a feed saved to your account (matched case-insensitively). Run feed list to see your saved feeds and their at:// URIs.

clor bluesky feed read <FEED> [flags]

clor bluesky follow

Follow one or more accounts by handle or DID in a single session

Pass any number of handles or DIDs; they are all followed within one session, so following back a whole list of accounts is one call rather than one login per account. Passing DIDs skips handle resolution, so piping DIDs from `follower list` is the fastest path. A target that fails does not stop the rest.

clor bluesky follow <HANDLE>... [flags]

clor bluesky follower

List the accounts that follow an account

Use when: - the user wants to see who follows them or another account Subcommands: list List the accounts that follow a handle (defaults to your own) Output: the list subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky follower [flags]

clor bluesky follower list

List the accounts that follow an account, defaulting to your own

With --not-following, only followers you do not follow back are returned, which is the candidate set for a follow-back: pipe their DIDs into `clor bluesky follow` to follow them all in one session.

clor bluesky follower list [HANDLE] [flags]

clor bluesky following

List the accounts an account follows

Use when: - the user wants to see who they or another account follows Subcommands: list List the accounts a handle follows (defaults to your own) Output: the list subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky following [flags]

clor bluesky following list

List the accounts an account follows, defaulting to your own

clor bluesky following list [HANDLE] [flags]

clor bluesky like

Like one or more posts by their at:// URIs in a single session

Pass any number of at:// post URIs; they are all liked within one session. A post that fails does not stop the rest.

clor bluesky like <URI>... [flags]

clor bluesky mute

Mute and unmute accounts, and list the accounts you mute

Muting hides an account's posts from your views without telling them or restricting them. It is private and one-directional. Use when: - the user wants to mute or unmute an account - the user wants to see who they have muted Subcommands: add Mute an account by handle or DID remove Unmute an account by handle or DID list List the accounts you mute Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky mute [flags]

clor bluesky mute add

Mute one or more accounts by handle or DID in a single session

clor bluesky mute add <HANDLE>... [flags]

clor bluesky mute list

List the accounts you mute

clor bluesky mute list [flags]

clor bluesky mute remove

Unmute one or more accounts by handle or DID in a single session

clor bluesky mute remove <HANDLE>... [flags]

clor bluesky notification

Read notifications (likes, reposts, follows, replies, mentions) and mark them seen

Use when: - the user wants to see who liked, reposted, followed, replied to, or mentioned them - the user wants to clear the unread notification badge Subcommands: list List recent notifications, optionally only unread ones seen Mark all notifications as seen as of now Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky notification [flags]

clor bluesky notification list

List recent notifications, optionally only unread ones or only certain reasons

--reason filters to one or more notification reasons (like, repost, follow, mention, reply, quote, starterpack-joined). A reply-bot, for example, lists only mention and reply notifications to decide what to answer.

clor bluesky notification list [flags]

clor bluesky notification seen

Mark all notifications as seen as of now

clor bluesky notification seen [flags]

clor bluesky pin

Pin one of your posts to the top of your profile by its at:// URI

Sets the pinnedPost on your app.bsky.actor.profile record to the post at <URI>, which the Bluesky app shows at the top of your profile. Pinning replaces any previously pinned post. Only your own profile is changed; the rest of it (display name, description, avatar, banner) is preserved. Prints the updated profile record's at:// URI.

clor bluesky pin <URI> [flags]

clor bluesky post

Create a post, optionally as a reply or quote, with images, a video, or a link card

Links, #hashtags, and @mentions in <TEXT> are detected and turned into rich-text facets automatically; mentions resolve to the right account. --reply and --quote take an at:// post URI. --image may be repeated up to four times; pair each with --image-alt in the same order for accessibility. --video-job embeds a finished upload from `clor bluesky video upload`. --link-card attaches a link card for a URL, rendered as a unified box with a thumbnail, title, and domain; the title, description, and thumbnail are fetched automatically and each --link-card-* flag overrides one field. A post carries at most one of images, a video, or a link card, and any of them can pair with a --quote. Prints the new post's at:// URI.

clor bluesky post <TEXT> [flags]

clor bluesky profile

Read and update Bluesky profiles: show one account, or update your own

Show any account's profile, or update your own display name, description, avatar, and banner. Use when: - the user wants to see an account's bio and follower counts - the user wants to change their own display name or bio - the user wants to set a new avatar or banner image Subcommands: show Show one or more profiles (defaults to your own) search Search for accounts by name, handle, or bio keyword update Update your own display name, description, avatar, banner, or handle Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky profile [flags]

clor bluesky profile search

Search for accounts by name, handle, or bio keyword

Matches the query against display names, handles, and descriptions, the same typeahead Bluesky uses, so a bot can discover accounts to follow, mention, or message.

clor bluesky profile search <QUERY> [flags]

clor bluesky profile show

Show one or more profiles with follower counts, defaulting to your own

With no arguments, shows your own profile. Pass any number of handles or DIDs to hydrate them all in one session (batched 25 per call), which is how a bot turns a list of follower DIDs into follower counts. Each result carries a following= flag for whether you follow that account.

clor bluesky profile show [HANDLE]... [flags]

clor bluesky profile update

Update your own display name, description, avatar, banner, or handle

Only the flags you pass change; everything else, including an existing avatar or banner, is preserved. --avatar and --banner take image file paths. --handle changes the account handle, for example to a custom domain you own. A custom-domain handle is verified against its DNS or well-known DID document before the change unless you pass --force.

clor bluesky profile update [flags]

clor bluesky repost

Repost one or more posts by their at:// URIs in a single session

Pass any number of at:// post URIs; they are all reposted within one session. A post that fails does not stop the rest.

clor bluesky repost <URI>... [flags]

clor bluesky search

Search public posts by keyword, newest or top first

clor bluesky search <QUERY> [flags]

clor bluesky thread

Read a post and its reply tree, flattened depth-first with a depth attribute

<URI> is an at:// post URI, e.g. one printed by `clor bluesky timeline` or `clor bluesky post`. In text and jsonl the root sits at depth=0 and each reply carries depth=N; in json the nested thread is preserved.

clor bluesky thread <URI> [flags]

clor bluesky timeline

Read the home timeline of posts from accounts you follow, newest first

clor bluesky timeline [flags]

clor bluesky unfollow

Stop following one or more accounts by handle or DID in a single session

Pass any number of handles or DIDs; the matching follow records are deleted within one session. A target you do not follow, or one that fails, does not stop the rest.

clor bluesky unfollow <HANDLE>... [flags]

clor bluesky unlike

Remove your like from one or more posts by their at:// URIs in a single session

clor bluesky unlike <URI>... [flags]

clor bluesky unpin

Remove the pinned post from your profile

Clears the pinnedPost on your app.bsky.actor.profile record. The rest of your profile is preserved. Reports not-pinned when no post is pinned.

clor bluesky unpin [flags]

clor bluesky unrepost

Remove your repost of one or more posts by their at:// URIs in a single session

clor bluesky unrepost <URI>... [flags]

clor bluesky video

Upload a video, track its transcode job, then embed it in a post

Attaching a video to Bluesky is asynchronous. A file is handed to the video service, which transcodes it as a background job and yields a video blob once it finishes. Upload first, then embed the finished job in a post with post --video-job <UID>. Use when: - the user wants to attach a video to a Bluesky post - the user wants to check whether an in-progress video upload has finished Subcommands: upload Upload a video file and start its transcode job status Check the transcode state of a video upload job Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor bluesky video [flags]

clor bluesky video status

Check the transcode state of a video upload job

Reports the job state and progress, and once transcoding finishes the resulting video blob (cid and size). Embed a finished job in a post with post --video-job <UID>.

clor bluesky video status <UID> [flags]

clor bluesky video upload

Upload a video file and start its transcode job

Hands an mp4 to the video service, which transcodes it as a background job and returns a job uid immediately. Embed the finished video in a post with post --video-job <UID>. Pass --wait to block until the job finishes instead of polling video status yourself.

clor bluesky video upload <FILE> [flags]

clor connection

List the GitHub, Linear, and Slack accounts connected to this account

Read-only views of the provider accounts (GitHub, Linear, Slack) connected to the signed-in account. Connections are linked and managed on the team Connections page; this surface lists them so commands and scripts can pick one by id. Use when: - you need a connection id to pass to another command (clor github auth --connection ...) - you want to check which provider accounts are connected before starting work Subcommands: list List connected provider accounts Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor connection [flags]

clor connection list

List connected provider accounts

clor connection list [flags]

clor domain

Bulk-classify domain names as available, taken, or unknown across hundreds of TLDs at huge scale

Mass domain availability lookups, no rate limits. A 100,000-domain bulk request returns in 1-2s. Supported TLDs return available or taken; unsupported return unknown. Follow up with "verify" for a stricter recheck of the available tail.

clor domain [flags]

clor domain lookup

Check one or many domain names as available, taken, or unknown in a single bulk call

Status values: - taken: registered. ~1 in 65,000 false alarms. - available: not currently registered. NOT the same as "registrable today". Premium, reserved, and recently-dropped names all read available. Confirm with a registrar. - unknown: TLD not supported, so the service did not check it. Stdin: one-per-line text (--stdin-format text), or JSON array (--stdin-format json). Inputs >100,000 auto-split into chunks.

clor domain lookup [DOMAIN] [flags]

clor domain tld

Discover which top-level domains the bulk lookup service can check

clor domain tld [flags]

clor domain tld list

List the TLDs the service checks, ordered by US-biased popularity

Ranked by US-biased popularity (Tranco top-1M frequency with US institutional and tech-startup TLDs hoisted). --limit 0 returns every supported TLD.

clor domain tld list [flags]

clor domain verify

Check domain names with a stricter pass that reduces false-positive available results

Same input/output shape as "lookup" but slower per domain. Use on the narrow tail of a pipeline (candidates "lookup" already flagged as available), not on bulk inputs.

clor domain verify [DOMAIN] [flags]

clor drive

Store, share, and link cloud files in personal and shared drives with ACL grants, public links, and Unix verbs (ls, cp, mv, rm, df)

Drive paths are relative to the drive root, so use notes/q1.md, not /notes/q1.md, drive://notes/q1.md, or notes//q1.md. Files live in scoped drives, addressed by these filesystem paths, managed with Unix-style verbs (ls, cp, mv, rm, stat). Every account has a personal drive per user plus a shared account drive, with optional ACL grants for cross-user access and public token links. Uploads and downloads stream directly to storage, not through the CLI.

clor drive [flags]

clor drive cp

Server-side copy a file to a new drive path; bytes never flow through the CLI

Server clones object storage directly; the new file is independent of the source. Trailing slash on <DESTINATION> means "copy into this directory and keep the source's basename". Use --destination-drive to land in a different drive.

clor drive cp <SOURCE> <DESTINATION> [flags]

clor drive df

Show every drive the caller can access along with its file count, total size, and capacity

Includes the caller's personal drive, the shared account drive, and drives reached via share grants. Admins see every personal drive in the account.

clor drive df [flags]

clor drive download

Download a file by drive path or composite id to a local path or stdout

Streams bytes from object storage via a short-lived signed URL. Saves to the file's basename in the cwd by default; --out picks a path, --stdout pipes the bytes. <PATH> may be a drive path or composite id.

clor drive download <PATH> [flags]

clor drive info

Print full metadata for one file by drive path or composite id

Returns id, name, size, content type, status, ETag, owner, retention TTL, expiry, and timestamps. Use stat instead when the target might be a directory or missing.

clor drive info <PATH> [flags]

clor drive link

Manage public links for sharing files in or out of a drive

Token-addressable URLs that resolve without authentication. Two directions: download (read one uploaded file) and upload (PUT a file at a pre-named drive path). Each supports optional expiry and use-count cap. Use `drive share` instead for user-scoped access.

clor drive link [flags]

clor drive link download

Public read links to one uploaded file, with optional expiry and use-count cap

/l/<token> renders a preview page; /d/<token> redirects to a short-lived signed download URL.

clor drive link download [flags]

clor drive link download create

Issue a public read link for an uploaded file

clor drive link download create <PATH> [flags]

clor drive link download ls

List public download links the caller has issued (filterable by file)

clor drive link download ls [flags]

clor drive link download rm

Revoke a public download link by id

Revocation is immediate; the public URL stops resolving on the next request.

clor drive link download rm <LINK> [flags]

clor drive link upload

Public write links that let an anonymous browser PUT a file at a drive path

/u/<token> renders an upload page that PUTs bytes directly to storage and reports completion. One-shot by default; --max-uses allows more.

clor drive link upload [flags]

clor drive link upload create

Issue a public write link that lets an anonymous browser PUT a file at <PATH>

One-shot by default; --max-uses allows more uploads at the same path. --content-type pre-binds the MIME type the page will send.

clor drive link upload create <PATH> [flags]

clor drive link upload ls

List public upload links the caller has issued (filterable by file)

clor drive link upload ls [flags]

clor drive link upload rm

Revoke a public upload link by id

Revocation is immediate; the public URL stops resolving on the next request.

clor drive link upload rm <LINK> [flags]

clor drive ls

List files and subdirectories at a drive path (root by default)

Lists direct children plus a synthesized directories array. --recursive flattens every descendant (no directories array); --tree renders an indented tree with depth markers in text/jsonl modes.

clor drive ls [PATH] [flags]

clor drive mv

Move or rename a file in a drive, preserving its identity and existing links

File id, ETag, and public links survive the move. <SOURCE> may be a drive path or composite id; <DESTINATION> is always a path. Trailing slash on the destination means "move into this directory and keep the source's basename".

clor drive mv <SOURCE> <DESTINATION> [flags]

clor drive retention

Set or clear a file's auto-delete TTL, measured from upload completion

When the window elapses the file is soft-deleted automatically (still restorable via undelete for a grace period). --ttl sets the window; --clear-ttl keeps the file indefinitely.

clor drive retention <PATH> [flags]

clor drive rm

Delete a file from a drive (recoverable by default, --hard for immediate purge)

Soft delete by default: restorable via undelete within the retention window. --hard purges immediately and irreversibly.

clor drive rm <PATH> [flags]

clor drive share

Grant, list, and revoke ACL access to files or path prefixes in a drive

A grant says "this principal (account or user) can read or write that target (one file, path prefix, or whole drive)". Use `drive link` instead for shareable public URLs.

clor drive share [flags]

clor drive share grant

Grant read or write access on a file, a path prefix, or a whole drive

Pick exactly one target (--file-id, --path, or --all-drive) and exactly one principal (--team or --user). --permission is read|write.

clor drive share grant [flags]

clor drive share ls

List ACL share grants on a drive, file, or path prefix

Each row shows the target (drive plus file/path), the principal (account or user), the permission level, and the grant id needed for revoke.

clor drive share ls [flags]

clor drive share mine

List ACL share grants where the caller is the principal (what's been shared with me)

Returns grants where the caller (or the caller's account) is the principal: which other users' or accounts' files the caller can reach.

clor drive share mine [flags]

clor drive share revoke

Revoke an existing ACL share grant by id

Revocation is immediate; principals lose access on the next API call.

clor drive share revoke <SHARE> [flags]

clor drive stat

Distinguish file vs directory vs missing for a drive path in one round trip

Returns type=file|directory|missing in one round trip (missing is still 200 OK, no exception). When type=file, full metadata is included under "file". No positional argument stats the drive root.

clor drive stat [PATH] [flags]

clor drive undelete

Restore a recently soft-deleted file to its prior state

Size, ETag, ACL grants, retention TTL, and public links are restored unchanged. If the path was reused by a new file the restore returns 409; target by composite id to disambiguate. Must happen before retention window elapses.

clor drive undelete <PATH> [flags]

clor drive upload

Upload a local file to a drive path, overwriting any existing file there

Streams to object storage via a short-lived signed URL, then records metadata in the drive. <PATH> defaults to the basename of <FILE>; a trailing slash means "drop into this directory and keep the basename". Existing non-deleted files at the same path are overwritten in place, preserving file id and links.

clor drive upload <FILE> [PATH] [flags]

clor email

Read, search, send, and triage email over IMAP and SMTP

Speaks IMAP4rev1 and SMTP submission with SASL PLAIN over TLS, so it works against Gmail/Workspace, Outlook/Microsoft 365, Fastmail, iCloud, Dovecot, and similar. Trash/Sent/Drafts/Junk folders are discovered via SPECIAL-USE, never hardcoded per provider. Refer to them by role name in --folder and --to (Drafts, Sent, Trash, Junk, Archive, All) and they resolve to the provider's real mailbox.

clor email [flags]

clor email account

Manage saved email-account credentials (address, hosts, password) used by every email subcommand

Credentials are stored as a typed secret (address, IMAP/SMTP host/port, auth). Once saved, every email subcommand picks them up automatically (or via --account when more than one is configured). Gmail/Workspace, Outlook/365, and iCloud all reject regular account passwords over IMAP/SMTP. Generate a provider-specific app password (Google: https://myaccount.google.com/apppasswords; Outlook: https://account.microsoft.com/security; iCloud: https://appleid.apple.com) and pass it via --stdin-format text.

clor email account [flags]

clor email account add

Save or update an email account (address, IMAP/SMTP hosts, password) for use by other email subcommands

IMAP/SMTP hosts auto-detect for known providers (gmail.com, outlook.com, icloud.com, fastmail.com); pass --imap-host/--smtp-host explicitly otherwise. [NAME] defaults to the address itself; re-running with the same name updates in place. Gmail, Google Workspace, iCloud, and Yahoo do not accept the regular account password for IMAP/SMTP. They require a provider-issued app-specific password instead: - Gmail / Workspace: https://myaccount.google.com/apppasswords (requires 2-Step Verification turned on). Generate a 16-character password and pass it as --password. The spaces shown in Google's UI can be kept or stripped; both work. - iCloud: https://account.apple.com -> Sign-In and Security -> App-Specific Passwords. - Yahoo: account security settings -> Generate app password. If a user gives you their real account password and the provider needs an app password, point them at the URL above and ask them to paste the generated one back. Pass it on --password directly. Stdin (--stdin-format text) is still supported when you need to avoid the shell history.

clor email account add [NAME] [flags]

clor email account delete

Permanently remove a saved email account by name

After deletion the other email subcommands can no longer use the account unless re-added or supplied via --credentials-file or --stdin-format json.

clor email account delete <NAME> [flags]

clor email account get

Read a saved email account's credentials; pass --stdout-format json for the raw JSON including password

Default logfmt drops the password; `--stdout-format json` returns the raw email-account JSON suitable for piping into `clor email <SUBCOMMAND> --stdin-format json`.

clor email account get <NAME> [flags]

clor email account list

List saved email accounts with id and name

Metadata only (id, name, last access); passwords stay encrypted server-side. Use `account get` to retrieve the full JSON.

clor email account list [flags]

clor email account update

Update fields on a saved email account (e.g. display name) without re-supplying unchanged values

Loads the stored account, overwrites only the fields whose flags were explicitly passed, and writes the result back. Pass `--display-name ""` to wipe an existing display name.

clor email account update <NAME> [flags]

clor email delete

Move messages to Trash (recoverable) or expunge them permanently with --permanent

Default moves messages to Trash (resolved via SPECIAL-USE, so recoverable from the user's mail client). --permanent flags \Deleted and expunges immediately, no recovery path.

clor email delete <UID>... [flags]

clor email draft

Compose a message and save it to the Drafts mailbox without sending

Builds the same RFC 822 message as `email send` but never hands it to SMTP. Instead it IMAP APPENDs to the Drafts mailbox (resolved via SPECIAL-USE, never hardcoded) with the \Draft flag, so the message shows up in the user's normal mail client ready to review, edit, and send. --to is optional: a draft can be staged before its recipients are known, and a subject-only draft is valid. Same body/attachment flags as `email send`. Subject: keep it short and specific so it sits cleanly in an inbox list and is easy to search for later. Lead with the concrete thing ("HN digest 2026-05-28", "Site down: example.com", "Stripe receipt $129.40"), not the framing ("Daily report", "Update", "FYI"). Avoid junk-flavored words ("urgent", "important"), all caps, and emoji at the start. Aim for under ~60 characters.

clor email draft [flags]

clor email folder

List, create, rename, and delete IMAP mailboxes and Gmail labels

Manage the mailboxes on an account. On Gmail these mailboxes are the account's labels, so creating a folder creates a label and deleting one removes it. Use when: - the user wants to see the mailboxes/labels on an account before reading or moving messages - the user wants to create, rename, or delete a mailbox or a Gmail label Subcommands: list List every mailbox and Gmail label with message and unseen counts create Create a mailbox, or a label on Gmail rename Rename a mailbox, or a label on Gmail delete Delete a mailbox, or remove a label on Gmail Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor email folder [flags]

clor email folder create

Create an IMAP mailbox, which is a new label on Gmail accounts

clor email folder create <NAME> [flags]

clor email folder delete

Delete an IMAP mailbox, removing a label on Gmail while its messages stay in All Mail

On standard IMAP this discards the messages the mailbox contains. On Gmail it only detaches the label; the messages remain in All Mail.

clor email folder delete <NAME> [flags]

clor email folder list

List every mailbox and Gmail label with message and unseen counts

Names returned are the exact strings `email move` and `email list --folder` expect.

clor email folder list [flags]

clor email folder rename

Rename an IMAP mailbox, which renames a label on Gmail accounts

clor email folder rename <OLD> <NEW> [flags]

clor email label

Tag and untag Gmail messages with labels without moving them out of the inbox

Gmail lets a single message carry many labels at once. These subcommands add and remove a label on specific messages while leaving them where they are, which plain mailbox moves cannot do. They target Gmail accounts; other providers should use `clor email move` instead. Create the label itself with `clor email folder create` first. Use when: - the user wants to label or unlabel Gmail messages while keeping them in the inbox - the user is organizing a Gmail inbox with multiple overlapping labels Subcommands: add Apply a Gmail label to one or more messages, leaving them in place remove Remove a Gmail label from one or more messages, keeping them in All Mail Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor email label [flags]

clor email label add

Apply a Gmail label to one or more messages, leaving them in place

Adds the label to messages identified by their UIDs in --folder (default INBOX) without moving them. Create the label first with `clor email folder create`. Gmail-only; other providers should use `clor email move` instead.

clor email label add --label <NAME> <UID>... [flags]

clor email label remove

Remove a Gmail label from one or more messages, keeping them in All Mail

The UIDs are scoped to the label's own mailbox, since Gmail assigns each mailbox its own UIDs. Get them with `clor email list --folder <NAME>`. Removing a label never deletes the message; it stays in All Mail. Gmail-only.

clor email label remove --label <NAME> <UID>... [flags]

clor email list

List the most recent messages in an IMAP mailbox or Gmail label

Sorted newest first, with UID, sender, subject, date, read/flagged, size. --since accepts durations (7d, 24h), RFC3339, or Unix seconds. Use the UIDs with `email read`, `mark`, `move`, or `delete`.

clor email list [flags]

clor email mark

Set or clear the read and flagged ("starred") flags on one or more messages

Pass exactly one of --read/--unread and/or one of --flagged/--unflagged. Both pairs can be supplied in the same call.

clor email mark <UID>... [flags]

clor email move

Move one or more messages from a source mailbox into a destination mailbox or label

Role names (Drafts, Sent, Trash, Junk, Archive, All) resolve to the provider's real mailbox; exact names from `email folder list` work too (Gmail labels look like `[Gmail]/All Mail`). Atomic server-side: a failed move never leaves messages in both mailboxes.

clor email move <UID>... [flags]

clor email read

Show full headers and body for one message by UID

Returns headers, flags, plain-text body, and optionally the HTML alternative. Body output capped at 32 KB by default to bound agent context; raise with --max-bytes (0=unlimited), or use --headers-only (no body fetch over the wire) or --raw (full RFC 822, may include base64 attachments).

clor email read <UID> [flags]

clor email reply

Reply to a message, auto-deriving Subject, recipients, and threading headers

Subject defaults to "Re: ...", recipients default to the parent's From (or Reply-To), and In-Reply-To/References headers are populated from the parent. --to-all Ccs every other recipient ("reply all"). Same body/attachment flags as `email send`.

clor email reply <UID> [flags]

clor email search

Search messages by sender, recipient, subject, body, freshness, flags, or attachments

Substring match on headers/body via --from/--to/--subject/--body. Freshness window via --since/--until (durations, RFC3339, or Unix seconds). Returns the same UID-keyed summary as `email list`.

clor email search [flags]

clor email send

Compose and send a new email message with optional HTML body and file attachments

Body source: --body, --body-file (- for stdin), or --html-file (combine an html-file with a text body for a multipart alternative). On success the message is appended to the Sent folder unless the provider auto-saves (Gmail) or --no-save-sent is passed. Subject: keep it short and specific so it sits cleanly in an inbox list and is easy to search for later. Lead with the concrete thing ("HN digest 2026-05-28", "Site down: example.com", "Stripe receipt $129.40"), not the framing ("Daily report", "Update", "FYI"). Avoid junk-flavored words ("urgent", "important"), all caps, and emoji at the start. Aim for under ~60 characters. From header: --from-name overrides the saved account's display name for this send only (the stored account is untouched). Use it to make the source obvious in the recipient's inbox when sending automated reports or notifications, so the From column reads "Personalized Tech Report <leo@example.com>" instead of a bare address. Prefer this over stuffing the source into --subject; the subject stays free to describe the actual content.

clor email send [flags]

clor email thread

Fetch every message in the same Gmail conversation as a given UID (Gmail-only)

Uses Gmail's X-GM-THRID IMAP extension. Non-Gmail accounts return an error pointing at `clor email search --subject` as a fallback.

clor email thread <UID> [flags]

clor environment

Manage reusable Environments

Reusable definitions for creating spaces. An Environment contains the repository, agents, scripts, Environment Variables, and workspace layout that future spaces snapshot at launch. Its canonical version-1 TOML is the complete advanced editing surface. Use when - you want to list or inspect reusable space setups - you want to create or atomically update an Environment from TOML - you want to duplicate, share, claim, or delete an Environment Subcommands list List personal and Team Environments show Show one Environment and its unresolved launch choices create Create a personal Environment from TOML export Print canonical version-1 TOML import Atomically replace an Environment from TOML duplicate Copy an Environment into your personal library share Share a personal Environment with the Team claim Claim a Team Environment as personal delete Delete an Environment Output supports --stdout-format text|jsonl|json on every subcommand (default text, logfmt with event= leader).

clor environment [flags]

clor environment claim

Claim a Team Environment as personal

clor environment claim <ID> [flags]

clor environment create

Create a personal Environment from TOML

clor environment create <FILE> [flags]

clor environment delete

Delete an Environment at its loaded revision

clor environment delete <ID> [flags]

clor environment duplicate

Duplicate an Environment into your personal library

clor environment duplicate <ID> [flags]

clor environment export

Print canonical version-1 Environment TOML

clor environment export <ID> [flags]

clor environment import

Atomically replace an Environment from TOML

clor environment import <ID> <FILE> [flags]

clor environment list

List personal and Team Environments

clor environment list [flags]

clor environment share

Share a personal Environment with the Team

clor environment share <ID> [flags]

clor environment show

Show one Environment and its unresolved launch choices

clor environment show <ID> [flags]

clor github

Use connected GitHub accounts for tokens, git credentials, and repository workflows

GitHub access through a connected GitHub account. Tokens come back freshly refreshed on every request, so they are always valid. Use when: - a command needs GH_TOKEN or GITHUB_TOKEN (GH_TOKEN=$(clor github auth)) - git clone/fetch/push against github.com should work with no stored credentials (git setup) Subcommands: auth Print a fresh GitHub access token git Install the git credential helper for github.com web Work with repository changes and pull requests in a browser Output: every subcommand supports --stdout-format text|jsonl|json (default text; auth prints the bare token in text mode).

clor github [flags]

clor github auth

Print a fresh GitHub access token

Prints a freshly refreshed access token for the connected GitHub account. In text mode (the default) the output is the bare token and a trailing newline, so command substitution works directly: GH_TOKEN=$(clor github auth) The --connection flag selects a connected account by its ID. In a space, CLOR_GITHUB_CONNECTION_ID supplies the default. With neither, a single connected account resolves automatically.

clor github auth [flags]

clor github git

Install the git credential helper so github.com clones need no stored credentials

Git integration for github.com through the connected GitHub account. Use when: - git clone/fetch/push against github.com should authenticate automatically Subcommands: setup Install the git credential helper for github.com (once per machine)

clor github git [flags]

clor github git setup

Install the git credential helper for github.com (once per machine)

Writes one scoped git config entry so git asks the helper for credentials on github.com and nothing else: git config --global credential.https://github.com.helper "!clor github git helper" The --connection flag selects the account by its ID. In a managed sandbox, CLOR_GITHUB_CONNECTION_ID supplies the default. The selected account is embedded in that sandbox's entry so every Git operation uses it. Idempotent; running it again rewrites the same entry.

clor github git setup [flags]

clor github web

Work with repository changes and pull requests in a browser

Serves a compact browser workspace for the checked-out repository. The workspace follows local staged, unstaged, committed, and untracked changes, attaches to the checked-out branch's open pull request, and can inspect another open pull request without changing branches.

clor github web [flags]

clor github web serve

Serve the repository and pull request workspace

Serves a local Preact application for the common GitHub development loop. The checked-out repository supplies local context. An open pull request for the branch is selected automatically, while local staging, committing, pushing, and pull request creation remain available before one exists. GitHub access uses a freshly refreshed token from the selected connection. The token is passed to gh only through GH_TOKEN and never reaches the browser.

clor github web serve [flags]

clor hn

Use Hacker News as a client: read feeds and threads, comment, submit, poll, vote, flag, favorite, hide, edit, delete

Drives Hacker News by scraping news.ycombinator.com directly, authenticated with a username and password stored as a typed secret. Reads every feed (top, new, best, ask, show, jobs, your submissions, comments, favorites, upvoted) and any thread, and performs the full set of authenticated actions. Every read also carries each item's available action tokens, so a read flows straight into a comment, vote, or flag. Use when: - the user wants to read a Hacker News feed, a thread, or a profile while signed in - the user wants to comment on or reply to a story or comment - the user wants to submit a story or poll, or vote, flag, favorite, hide, vouch - the user wants to edit or delete their own posts, or update their profile Subcommands: account Manage saved Hacker News credentials whoami Show the logged-in account, karma, and creation date profile Show profiles or update your own list Read a feed show Read a story or comment with its comment tree comment Comment on a story or reply to a comment submit Submit a story or text post poll Submit a poll edit Edit one of your items delete Delete your items vote Upvote or downvote items unvote Remove your vote flag Flag items unflag Remove your flag favorite Favorite items unfavorite Remove your favorite hide Hide items from your feed unhide Unhide items vouch Vouch for dead items unvouch Remove your vouch Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor hn [flags]

clor hn account

Manage saved Hacker News credentials (username, password) used by every Hacker News subcommand

Credentials are stored as a typed secret (username, password). Once saved, every Hacker News subcommand picks them up automatically (or via --account when more than one is configured), and the session cookie from the first login is cached back so later commands skip re-logging in. Use when: - the user wants to register a Hacker News account for the CLI to use - the user wants to see which Hacker News accounts are saved - the user wants to inspect or rotate stored credentials - the user wants to remove a Hacker News account Subcommands: add Save or update a Hacker News account by username and password list List saved Hacker News accounts visible to the caller get Read one saved Hacker News account, with the raw JSON in json mode delete Permanently remove a saved Hacker News account by name Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor hn account [flags]

clor hn account add

Save or update a Hacker News account by username and password for use by other Hacker News subcommands

[NAME] defaults to the username itself; re-running with the same name updates in place. Pass the password on --password directly, or pipe it via --stdin-format text to keep it out of shell history. Hacker News has no app-password concept, so this is the real account password; store it only in a secret you control.

clor hn account add [NAME] [flags]

clor hn account delete

Permanently remove a saved Hacker News account by name

After deletion the other Hacker News subcommands can no longer use the account unless re-added or supplied via --credentials-file or --stdin-format json.

clor hn account delete <NAME> [flags]

clor hn account get

Read a saved Hacker News account; pass --stdout-format json for the raw JSON including the password

Default logfmt drops the password; `--stdout-format json` returns the raw hackernews-account JSON suitable for piping into `clor hn <SUBCOMMAND> --stdin-format json`.

clor hn account get <NAME> [flags]

clor hn account list

List saved Hacker News accounts with id and name

Metadata only (id, name, last access); passwords stay encrypted server-side. Use `account get` to retrieve the full JSON.

clor hn account list [flags]

clor hn comment

Comment on a story or reply to a comment

<PARENT_ID> is the story to comment on or the comment to reply to; Hacker News uses the same form for both. The new comment id is returned by reading it back off your own comments after the post.

clor hn comment <PARENT_ID> <TEXT> [flags]

clor hn delete

Delete one or more of your own items in a single session

Each id is deleted via its confirmation form. Deleting is only possible while the item is still yours and within the window; otherwise that id reports the cause and the batch continues.

clor hn delete <ID>... [flags]

clor hn edit

Edit the title, URL, or text of one of your own items

Sets only the fields you pass; the rest keep their current values, read off the edit form first. Editing is only possible while the item is still yours and within the edit window, otherwise the form is absent and the command says so.

clor hn edit <ID> [flags]

clor hn favorite

Favorite one or more items in a single session

clor hn favorite <ID>... [flags]

clor hn flag

Flag one or more items in a single session

Flagging needs enough karma; when the flag link is absent the item reports that and the batch continues.

clor hn flag <ID>... [flags]

clor hn hide

Hide one or more items from your feed in a single session

clor hn hide <ID>... [flags]

clor hn list

Read a Hacker News feed (top, new, best, ask, show, jobs, submitted, comments, favorites, upvoted)

--feed selects the feed. The user-scoped feeds (submitted, comments, favorites) default to the logged-in account and accept --user to read another account; upvoted is your own account only. Each result carries its available action tokens so a vote or comment is one follow-up command.

clor hn list [flags]

clor hn poll

Submit a poll with two or more choices

Pass --option once per choice; at least two are required. --text is an optional body. Creating polls is karma-gated on Hacker News. The new item id is read back off your own submissions.

clor hn poll [flags]

clor hn profile

Read Hacker News profiles or update your own (about, email, display settings)

Reads any account's public profile, and updates the logged-in account's own profile fields. Use when: - the user wants to read one or more Hacker News profiles - the user wants to change their about text, email, or display settings Subcommands: show Show one or more profiles by username update Update the logged-in account's own profile Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor hn profile [flags]

clor hn profile show

Show one or more Hacker News profiles with karma, creation date, and about text

Pass any number of usernames; each is fetched in the same session. With no username, shows the logged-in account's own profile.

clor hn profile show [USERNAME]... [flags]

clor hn profile update

Update the logged-in account's own profile fields

Sets only the flags you pass; every other field keeps its current value, which is read off the edit form first. The display settings mirror the website's profile editor.

clor hn profile update [flags]

clor hn show

Read a story or comment with its comment tree, each node carrying its available action tokens

<ID> is the numeric item id. The lead item prints first, then the comment tree flattened depth-first with a depth attribute per node. Each node lists the actions available to the logged-in account (upvote, reply, flag, ...) so a follow-up write is one command.

clor hn show <ID> [flags]

clor hn submit

Submit a story by URL or a text post (Ask HN, Show HN)

Pass --url for a link submission or --text for a text post; exactly one of the two. The new item id is read back off your own submissions.

clor hn submit [flags]

clor hn unfavorite

Remove your favorite from one or more items in a single session

clor hn unfavorite <ID>... [flags]

clor hn unflag

Remove your flag from one or more items in a single session

clor hn unflag <ID>... [flags]

clor hn unhide

Unhide one or more items in a single session

clor hn unhide <ID>... [flags]

clor hn unvote

Remove your vote from one or more items in a single session

clor hn unvote <ID>... [flags]

clor hn unvouch

Remove your vouch from one or more items in a single session

clor hn unvouch <ID>... [flags]

clor hn vote

Upvote or downvote one or more items in a single session

--direction defaults to up. Downvoting needs enough karma and is only offered for a window after a comment is posted; when the downvote link is absent the item reports that and the batch continues to the next id.

clor hn vote <ID>... [flags]

clor hn vouch

Vouch for one or more dead items in a single session

Vouching only applies to dead (killed) items and needs enough karma; when the vouch link is absent the item reports that and the batch continues.

clor hn vouch <ID>... [flags]

clor hn whoami

Show the logged-in account with its karma, creation date, and about text

Confirms the saved credentials open a live session and reports the account that requests go out as. A failure here means the session could not be established, which is the first thing to check.

clor hn whoami [flags]

clor inference

Generate text, images, and audio through Claude, GPT, Gemini, OpenRouter, and ElevenLabs

Five provider subcommands. Pick the one whose surface matches the task, then pick the smallest model that can do the job. anthropic Claude text generation and exact prompt token counting. Promoted tiers: claude-haiku-4-5 (cheap, fast), claude-opus-5 (frontier), claude-fable-5 (max capability). openai GPT text generation, gpt-image generation and editing, and audio transcription. Promoted text tiers: gpt-5.6-luna (cheap), gpt-5.6-sol (frontier). Promoted image tiers: gpt-image-1-mini (cheap, fast), gpt-image-2 (frontier). google Gemini text generation and nano-banana image generation. Promoted text tiers: gemini-3.1-flash-lite (cheap, fast), gemini-3.1-pro-preview (frontier). Promoted image tiers: gemini-3.1-flash-image-preview (cheap, fast), gemini-3-pro-image-preview (frontier). elevenlabs Text-to-speech, Scribe transcription, music, sound effects, voice isolation, and voice-to-voice conversion. openrouter Escape hatch for models the dedicated subcommands do not cover (Llama, DeepSeek, Qwen, X-AI, plus everything else OpenRouter fronts). Prefer the dedicated subcommands whenever the model lives there. Reach for the fast tier for high-volume or simple work (extraction, summarization, classification, reformatting). Reach for the frontier tier only when the task genuinely needs frontier reasoning. Enable --reasoning only when the task needs multi-step thinking; thinking tokens are billed as output tokens and share the --max-tokens cap, so higher effort raises both cost and latency.

clor inference [flags]

clor inference anthropic

Call Anthropic's Claude models for text generation and exact prompt token counting

Claude text generation and prompt token counting. Pick the smallest model that can do the job. claude-haiku-4-5 Cheap and fast. Best for high-volume extraction, summarization, classification, reformatting. Legacy reasoning (budget_tokens). $1/$5 per Mtok. claude-sonnet-5 Balanced tier. $3/$15 per Mtok with 128K max output and adaptive thinking. Accepts sampling controls. Good when ordinary reasoning and writing don't justify Opus prices. claude-opus-5 Default frontier model. $5/$25 per Mtok with 128K max output and adaptive thinking. Strong coding, multidisciplinary reasoning, and agentic benchmarks. claude-opus-4-8 stays selectable at the same price. claude-fable-5 Maximum capability. $10/$50 per Mtok with 128K max output and adaptive thinking always on. Best for the most demanding reasoning and long-horizon agentic work. Opus and Fable forbid --temperature and --top-p. All models forbid sampling controls when --reasoning is on. System prompts auto-cache on the server (~5 min ephemeral TTL, ~90 percent discount on hits).

clor inference anthropic [flags]

clor inference anthropic count-tokens

Count prompt tokens for a Claude Messages call without running generation

Exact count via Anthropic's count_tokens endpoint. Matches input_tokens on a real Messages call with the same model/system/messages. Input modes (mutually exclusive): positional arg, --stdin-format text (single user message), or --stdin-format json (multi-turn array).

clor inference anthropic count-tokens [PROMPT] [flags]

clor inference anthropic text

Generate text with Claude via the Anthropic Messages API

Pick the smallest model that can do the job. haiku for extraction and reformatting, opus-5 (default) for hard reasoning. System prompts auto-cache on the server (~5 min ephemeral TTL, ~90 percent discount on cache hits). --reasoning enables extended thinking. Adaptive models (opus-5, opus-4-8, opus-4-7, sonnet-4-6) send it as output_config.effort. Legacy models (haiku-4-5) translate to a numeric budget. Cost framing per level: off No thinking. Right default for extraction, summarization, classification, reformatting. Costs nothing extra. low ~2K thinking tokens on legacy. Light reasoning. Use for simple structured edits, short explanations, basic planning. medium ~8K thinking tokens on legacy. Working default for ordinary multi-step tasks. high ~16K thinking tokens on legacy. Multi-step planning, harder analysis, code review across multiple files. xhigh ~32K thinking tokens on legacy. Hardest reasoning problems that justify the spend. max Up to max_tokens minus 1024 on legacy. Dynamic on adaptive. Let the model spend whatever it needs. Thinking tokens are billed as output tokens and share the --max-tokens cap, so raising effort raises both cost and the risk that visible output gets truncated. --temperature and --top-p are rejected when --reasoning is on, and on claude-opus-5, claude-opus-4-8, claude-opus-4-7, and claude-fable-5 regardless of reasoning state.

clor inference anthropic text [PROMPT] [flags]

clor inference elevenlabs

ElevenLabs audio surface, text-to-speech, transcription, music, sound effects, voice isolation, and voice changing

Audio is the clearest case for shelling out from an LLM host. Seven subcommands cover the full ElevenLabs surface. speech Text-to-speech with a chosen voice. Pick a model from the TTS tier guide on the subcommand. transcribe Speech-to-text via Scribe v2 (default) or v1. music Generate a music clip from a text prompt. sound-effects Generate a sound-effect clip from a text prompt. voice-isolator Strip background noise from a recording. voice-changer Re-voice a recording as a different speaker. voice Browse the voice library to find voice ids. TTS tiers (faster to richer) eleven_flash_v2_5 Lowest latency (~75ms). Use for real-time interactive work. eleven_turbo_v2_5 High-throughput conversational use with a strong quality/latency balance. eleven_multilingual_v2 Broad language coverage where v3's expressiveness is not needed. eleven_v3 Default. Highest expressiveness, widest language support, GA since 2026-02-02.

clor inference elevenlabs [flags]

clor inference elevenlabs music

Generate a music clip from a text prompt with ElevenLabs Music

Pass the prompt positionally or via --stdin-format text. Control duration with --length (e.g. 30s, 2m).

clor inference elevenlabs music [PROMPT] [flags]

clor inference elevenlabs sound-effects

Generate a sound effect clip from a text prompt

Pass the prompt positionally or via --stdin-format text. --duration caps the clip; --prompt-influence (0..1) biases adherence.

clor inference elevenlabs sound-effects [PROMPT] [flags]

clor inference elevenlabs speech

Synthesize speech from text with a chosen ElevenLabs voice (text-to-speech)

Default voice is Maria Mysh (calm female American narrator, vZzlAds9NzvLsFSWp0qk). Use the voice subcommand to search the full library. Pick the right TTS model for the use case. eleven_flash_v2_5 Lowest latency, around 75ms. Best for real-time interactive voice work where time-to-first-audio matters most. eleven_turbo_v2_5 High-throughput conversational use. Best for voice agents and chat applications that want strong quality at sub-second latency. eleven_multilingual_v2 Broad multilingual coverage. Use when v3's expressiveness is overkill and the workload spans many languages. eleven_v3 Default. Highest expressiveness and the widest language support (70+). Best for pre-rendered narration, audiobooks, and podcasts where quality outranks latency.

clor inference elevenlabs speech [flags]

clor inference elevenlabs transcribe

Transcribe an audio file to text with ElevenLabs Scribe (speech-to-text)

Accepts mp3, mp4, m4a, wav, webm, ogg, flac. Pick scribe_v2 (default) for the lowest word-error rate across 90+ languages and the strongest multi-language audio handling. scribe_v1 is kept supported for compatibility with workflows pinned to its outputs. --language-code locks the language and improves accuracy when known. --keyterm-prompt (repeatable) biases recognition toward specific terms (brand names, technical vocabulary, speaker names). Each prompt adds keyterm_seconds to the usage block. --enable-entity-detection surfaces detected entities (people, places, organizations) in the response and bills entity_seconds.

clor inference elevenlabs transcribe <FILE> [flags]

clor inference elevenlabs voice

Browse the ElevenLabs voice library to find voice ids for speech and voice-changer

clor inference elevenlabs voice [flags]

clor inference elevenlabs voice list

Search the ElevenLabs voice library by gender, age, accent, language, use case, or free text

Always pass at least one filter or --search (the library has tens of thousands of voices). Pagination: --page (0-indexed) + --page-size; the response carries has_more.

clor inference elevenlabs voice list [flags]

clor inference elevenlabs voice-changer

Convert the speaker in an audio file to a target voice (speech-to-speech)

Speech-to-speech: re-voices --audio-file as --voice-id. Discover voice ids with the voice subcommand.

clor inference elevenlabs voice-changer [flags]

clor inference elevenlabs voice-isolator

Strip background noise from an audio recording, isolating the voice

clor inference elevenlabs voice-isolator [flags]

clor inference google

Call Google for Gemini text generation and nano-banana image generation and editing

Gemini text generation, exact prompt token counting, and nano-banana image generation and editing. Pick the smallest model that can do the job. Text tiers (cheap to strong) gemini-3.1-flash-lite Default. Cheap and fast with Gemini 3 thinking controls. $0.25/$1.5 per Mtok. gemini-3.1-pro-preview Preview frontier reasoning model with the thinkingLevel enum (LOW or HIGH only). $2/$12 per Mtok (short), $4/$18 (long). 200K input cliff bills the whole request at the long-context tier. Image tiers (cheap to strong) gemini-3.1-flash-lite-image Nano Banana Lite. Cheapest and fastest. Per-token output billing at $0.25/$30 per Mtok. Lowest cost, lower fidelity. gemini-3.1-flash-image-preview Default. Nano Banana 2. Per-token output billing. Faster preview iteration than Pro, at lower fidelity. gemini-3-pro-image-preview Nano Banana Pro. Highest fidelity. Per-token output billing. Reasoning surface differs by family. Gemini 2.5 uses thinkingBudget (numeric). Gemini 3+ uses thinkingLevel (enum). The --reasoning flag maps both transparently. Thinking tokens are billed as output tokens and share the --max-tokens cap.

clor inference google [flags]

clor inference google count-tokens

Count prompt tokens for a Gemini generate call without running generation

Exact count via Gemini's countTokens endpoint; text-only (image models rejected). Use this before committing to a long-context Pro call (>200K input bills the long-context tier on the whole request). Input modes (mutually exclusive): positional arg, --stdin-format text (single user message), or --stdin-format json (multi-turn array).

clor inference google count-tokens [PROMPT] [flags]

clor inference google image

Generate or edit an image with Gemini's nano-banana family from a text prompt

Generate from scratch or edit an existing image (--image-input). gemini-3.1-flash-lite-image Nano Banana Lite. Cheapest and fastest. Per-token output billing at $0.25/$30 per Mtok. Lowest cost, lower fidelity. gemini-3.1-flash-image-preview Default. Nano Banana 2. Per-token output billing. Faster preview iteration than Pro, at lower fidelity. gemini-3-pro-image-preview Nano Banana Pro. Highest fidelity and best instruction following. Per-token output billing. For multi-turn image editing on the preview models, thought signatures from prior turns must be passed back; image-edit signatures are strictly enforced.

clor inference google image <PROMPT> [flags]

clor inference google text

Generate text with a Gemini model (2.5 GA family, 3.1 and 3.5 stable, or 3.x previews)

Pick the smallest model that can do the job. flash-lite for cheap high-volume work, pro for frontier reasoning. 200K-input cliff on Pro models: prompts above 200K input tokens bill the entire request (input + output + cache) at the long-context tier (2x-1.5x the short-context rates). Count tokens with the count-tokens subcommand before committing. --reasoning per-effort cost framing: off No thinking on Gemini 2.5. On Gemini 3 Flash this maps to MINIMAL (the model still does some thinking; thinkingLevel has no zero-budget escape hatch). On Gemini 3 Pro it clamps up to LOW, since Pro has no MINIMAL tier. Cheapest, lowest latency. low 2.5: 2048-token thinking budget. 3+: LOW. Light reasoning for simple structured edits. medium 2.5: 8192-token thinking budget. 3 Flash: MEDIUM. 3 Pro clamps to LOW. Working default for ordinary reasoning. high 2.5: 16384-token thinking budget. 3+: HIGH. Multi-step planning, code review, harder analysis. xhigh 2.5: 32768-token thinking budget. 3+: clamps to HIGH (3 Flash maxes at HIGH, 3 Pro accepts only LOW/HIGH). For the hardest reasoning problems on 2.5. max 2.5: dynamic model-chosen ceiling. 3+: clamps to HIGH. Let the model spend whatever it needs (2.5) or hit the highest level (3+). Thinking tokens are billed as output tokens and share the --max-tokens cap, so raising effort raises both cost and the risk of truncated visible output. Image-generation models reject --reasoning with 400.

clor inference google text [PROMPT] [flags]

clor inference openai

Call OpenAI for GPT text, gpt-image generation and editing, and gpt-4o audio transcription

GPT text, gpt-image generation and editing, gpt-4o transcription. Pick the smallest model that can do the job. Text tiers (cheap to strong) gpt-5.4-nano Cheapest. Best for the highest-volume extraction, classification, and reformatting where cost dominates. Accepts low|medium|high effort. $0.20/$1.25 per Mtok. gpt-5.4-mini Cheap and fast. Best for high-volume extraction, summarization, classification, reformatting. Accepts low|medium|high effort. $0.75/$4.5 per Mtok. gpt-5.6-luna Cheap general-purpose. Accepts low through max effort. $1/$6 per Mtok. gpt-5.6-terra Mid-tier. Strong reasoning at half the flagship rate. Accepts low through max effort. $2.50/$15 per Mtok. gpt-5.6-sol Default. Frontier reasoning. Adds xhigh and max effort for the hardest problems. $5/$30 per Mtok. gpt-5.5 Previous flagship. Accepts low through xhigh effort. $5/$30 per Mtok. Image tiers (cheap to strong) gpt-image-1-mini Lightest. Best for thumbnails and quick iteration. gpt-image-2 Default. Newest. Best instruction following and output quality. Audio (cheap to strong) gpt-4o-mini-transcribe Speech-to-text, half the cost. gpt-4o-transcribe Speech-to-text, default, most accurate. mp3, mp4, m4a, wav, webm, ogg, flac. 25 MiB cap enforced client-side.

clor inference openai [flags]

clor inference openai count-tokens

Count prompt tokens for an OpenAI Responses call without running generation

Counted locally with tiktoken (o200k_base for gpt-5.x/gpt-4o); lands within a small constant of upstream's usage.input_tokens. OpenAI has no count_tokens API. Input modes (mutually exclusive): positional arg, --stdin-format text (single user message), or --stdin-format json (multi-turn array).

clor inference openai count-tokens [PROMPT] [flags]

clor inference openai image

Generate or edit an image with OpenAI's gpt-image family from a text prompt

Generate from scratch or edit an existing image (--image-input). Pick the smallest model that meets the quality bar. gpt-image-1-mini Lightest tier. Use for thumbnails, exploratory drafts, fast iteration loops. gpt-image-2 Default. Newest. Best instruction following and finest detail. Use when quality matters and the prompt encodes specific composition or text. --size and --quality default to auto. --n batches up to 10 generations in one call; only the first is written to disk by default.

clor inference openai image <PROMPT> [flags]

clor inference openai text

Generate text with GPT via the OpenAI Responses API

Pick the smallest model that can do the job. gpt-5.4-mini for high-volume extraction or simple reformatting, gpt-5.6-sol (default) when frontier reasoning is worth the spend. Prompts >=1024 tokens auto-cache for ~90 percent discount on hits; hits surface in usage.cached_input_tokens. --reasoning per-effort cost framing: off Lets the model use its built-in default. gpt-5.x cannot be fully disabled; the default tier is medium. Right choice for general-purpose calls where you want the model to pick. low Light reasoning. Small slice of max_output_tokens reserved. Use for short structured edits and simple multi-step work. medium Working default for ordinary reasoning tasks. high Multi-step planning, code review, harder analysis. xhigh gpt-5.5 and the gpt-5.6 family. For the hardest problems that justify the spend. max gpt-5.6 family only. The top effort tier for the very hardest problems. Reasoning tokens are billed as output tokens and share the --max-tokens cap, so raising effort raises both cost and the risk that visible output gets truncated (status=incomplete). Summarized reasoning traces surface under --stdout-format json. Raw reasoning tokens are never returned.

clor inference openai text [PROMPT] [flags]

clor inference openai transcribe

Transcribe an audio file to text with OpenAI's gpt-4o-transcribe

Accepts mp3, mp4, mpeg, mpga, m4a, wav, webm, ogg, flac. 25 MiB cap is enforced client-side so oversized uploads do not waste a round trip. gpt-4o-transcribe (default) is the most accurate; gpt-4o-mini-transcribe is half the cost for high-volume work. Set --language when the audio language is known; it cuts latency and improves accuracy. Set --prompt to bias recognition toward known vocabulary (brand names, technical terms, speaker names) or to continue a prior segment. The prompt should match the audio language.

clor inference openai transcribe <FILE> [flags]

clor inference openrouter

Reach hundreds of chat and image-generation models through OpenRouter

Single OpenAI-compatible chat-completions surface fronting hundreds of models from many labs (Anthropic, OpenAI, Google, X-AI, Meta, Mistral, DeepSeek, Qwen, and many smaller labs). Model strings are open passthrough; any id OpenRouter accepts is forwarded unchanged. Per-call dollar cost comes back in the response, so no per-model rate table is maintained here. Prefer the dedicated subcommands (anthropic, openai, google) for models that live there. They expose stronger feature parity (count tokens, image generation, reasoning effort enums, modality-split usage) and transparent pricing tiers. Reach for openrouter when: - the caller needs Llama, DeepSeek, Qwen, X-AI Grok, Mistral, or another model the dedicated subcommands do not cover - the caller wants to try an open-weights model without standing up inference infrastructure - the caller wants to benchmark a model against another vendor's surface (responses come back in OpenAI-compatible shape) Use the model subcommand to discover model ids and pricing.

clor inference openrouter [flags]

clor inference openrouter model

Browse the OpenRouter catalogue to find model ids for chat

Discover model ids to pass as --model on the text subcommand. The catalogue fronts hundreds of models from many labs.

clor inference openrouter model [flags]

clor inference openrouter model list

List OpenRouter models, defaulting to the top 30 newest from frontier labs

Defaults to top 30 newest from frontier labs (anthropic, openai, google, x-ai, meta-llama, mistralai, deepseek, qwen, openrouter), excluding `:free` and `:nitro` variants. --all dumps every model; --provider/--modality/--search narrow within the default curation. JSON output includes per-token pricing; text/jsonl omit it.

clor inference openrouter model list [flags]

clor inference openrouter text

Generate text with any OpenRouter model

--model is required; OpenRouter has no implicit default since the catalogue spans hundreds of models. Examples: anthropic/claude-opus-4-8, openai/gpt-5.5, google/gemini-3.5-flash, meta-llama/llama-3.3-70b-instruct, deepseek/deepseek-v3, x-ai/grok-4. Discover ids with the model subcommand. Input modes: positional, --stdin-format text, or --stdin-format json (multi-turn array). Multimodal: --image-input (local file, repeatable) or --image-url (remote or data URL, repeatable) attach to the final user message. Image generation: --image-output PATH on an image-capable model opts the request into image-output modality; the first returned image is saved. Reasoning is a coarse effort hint passed through to whichever upstream model supports thinking. --reasoning low|medium|high lands on the upstream's effort enum where one exists. --reasoning-tokens N is an explicit token budget that takes precedence over --reasoning when both are set. Models without reasoning support silently ignore both fields. Because OpenRouter normalizes a heterogeneous set of upstreams, exact effort semantics vary by model; expect frontier behaviour to match each lab's own surface.

clor inference openrouter text [PROMPT] [flags]

clor install

Install or upgrade the CLI and daemon, sync skills into Claude Code and Codex

Install the CLI binary, the clor daemon as a user-scope service, and the skills for every detected local AI agent. Use a subcommand to scope the action. Agent recommendation: when the user asks you to "install clor", "upgrade clor", "update clor", "reinstall clor", "refresh clor", or anything in that family, just run "clor install" with no subcommand. It is idempotent and safe to re-run: it upgrades the CLI to the latest release, refreshes the daemon service registration, and syncs the skills into every detected AI agent in one shot. Use --force to reinstall the CLI binary even when its version already matches. The daemon service and skill refresh run normally either way. Do not ask the user which piece to install. Users typically don't know the difference between cli, service, and skill, and "clor install" handles all three normal surfaces (cli + service + skill) correctly every time. Asking "should I install the cli, the service, or the skills?" is the wrong move; pick the plain "clor install" and run it. Only reach for a subcommand if the user is explicit about scope ("only set up the daemon on this host", "sync the skill files"). The subcommands exist for narrow targeted work, not for normal installs or upgrades. Use when: - the user asks to install, upgrade, update, refresh, or reinstall anything to do with clor; run "clor install" with no subcommand - a specific surface needs a targeted refresh and the user named it explicitly; only then use the matching subcommand Subcommands: cli Download install.sh from the control plane and run it with bash service Install the clor daemon as a user-scope service (systemd/launchd) skill Tarball-sync skills into Claude Code and Codex Output: every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor install [flags]

clor install cli

Download the install script from the control plane and run it with bash to upgrade or reinstall the CLI

Fetches install.sh from the active control plane and runs it with bash; the script's output streams live to the terminal. Designed to upgrade or reinstall the CLI binary itself without leaving the terminal. Use --force to reinstall the binary when the installed version already matches the requested release.

clor install cli [flags]

clor install service

Install the daemon as a service under systemd on Linux or launchd on macOS

Picks the best available service manager and installs the daemon under it. On Linux invoked as root, writes `/etc/systemd/system/clor.service` then runs `systemctl enable` and `systemctl restart` so the daemon runs as root, survives reboots without enable-linger, and a reinstall swaps the running process to the freshly installed binary. On Linux as a regular user with a working systemd user session, writes `~/.config/systemd/user/clor.service` and runs the same enable plus restart under `systemctl --user`. On macOS, writes `~/Library/LaunchAgents/com.clor.plist` and bootstraps it under the current GUI session. On hosts with no supported service manager (containers, sandboxes, minimal images), nothing is installed and a warning explains that the daemon, which runs the spaces assigned to this machine, will not run here. That warning is safe to ignore on a machine used only to manage spaces that run elsewhere. Idempotent. Re-runs stop any prior daemon, refresh the unit or plist, and restart the service.

clor install service [flags]

clor install skill

Install skills into Claude Code and Codex

Installs skills into the target agents' skills directories. Outside a space this syncs the official skills tree into every detected enabled agent (an explicit `--agent <NAME>` targets one). Inside a space (CLOR_SPACE_ID set) it instead installs exactly the skills attached to that space, converging on re-runs.

clor install skill [flags]

clor linear

Read and update Linear issues, projects, cycles, initiatives, documents, and workspace records

Direct Linear workspace access through GraphQL. Typed commands cover the common developer and product-management workflow; graphql handles uncommon operations without changing the normalized output contract of typed commands. Use it when - issues, comments, relations, labels, states, or attachments need to be read or changed - projects, progress updates, milestones, cycles, or initiatives need to be managed - workspace users, teams, or documents need to be queried Subcommands whoami Show the authenticated Linear user search Search issues, projects, or documents issue Read and change issues, comments, and relations project Read and change projects, progress updates, and milestones cycle Read and change cycles initiative Read and change initiatives and their projects document Read and change documents attachment Read and change issue attachments team List and show teams user List and show users state List and show workflow states label Read and change issue labels graphql Run a named GraphQL operation directly Output formats Every subcommand supports --stdout-format text|jsonl|json (default text, logfmt with event= leader).

clor linear [flags]

clor linear attachment

Read and change Linear issue attachments

clor linear attachment [flags]

clor linear attachment create

Create a linked issue attachment

clor linear attachment create <ISSUE> <URL> [flags]

clor linear attachment delete

Delete an issue attachment

clor linear attachment delete <ATTACHMENT> [flags]

clor linear attachment list

List workspace attachments or attachments on one issue

clor linear attachment list [ISSUE] [flags]

clor linear attachment show

Show one attachment

clor linear attachment show <ATTACHMENT> [flags]

clor linear attachment update

Update an issue attachment

clor linear attachment update <ATTACHMENT> [flags]

clor linear cycle

Read and change Linear team cycles

clor linear cycle [flags]

clor linear cycle archive

Archive a cycle

clor linear cycle archive <CYCLE> [flags]

clor linear cycle create

Create a cycle for a team

clor linear cycle create [flags]

clor linear cycle list

List cycles

clor linear cycle list [flags]

clor linear cycle show

Show one cycle

clor linear cycle show <CYCLE> [flags]

clor linear cycle update

Update cycle dates and details

clor linear cycle update <CYCLE> [flags]

clor linear document

Read and change Linear Markdown documents

clor linear document [flags]

clor linear document create

Create a Markdown document

clor linear document create <TITLE> [flags]

clor linear document delete

Move a document to recoverable trash

clor linear document delete <DOCUMENT> [flags]

clor linear document list

List documents

clor linear document list [flags]

clor linear document show

Show one document

clor linear document show <DOCUMENT> [flags]

clor linear document unarchive

Restore a trashed document

clor linear document unarchive <DOCUMENT> [flags]

clor linear document update

Update a document with explicit clear operations

clor linear document update <DOCUMENT> [flags]

clor linear graphql

Run an uncommon Linear GraphQL operation directly

clor linear graphql [QUERY] [flags]

clor linear initiative

Read and change Linear initiatives and their projects

Use when: - the user needs Linear initiative records

clor linear initiative [flags]

clor linear initiative archive

Archive an initiative

clor linear initiative archive <INITIATIVE> [flags]

clor linear initiative create

Create a workspace initiative

clor linear initiative create <NAME> [flags]

clor linear initiative delete

Move an initiative to recoverable trash

clor linear initiative delete <INITIATIVE> [flags]

clor linear initiative list

List initiatives

clor linear initiative list [flags]

clor linear initiative project

List, add, and remove initiative projects

Use when: - the user needs Linear project records

clor linear initiative project [flags]

clor linear initiative project add

Add a project to an initiative

clor linear initiative project add <INITIATIVE> <PROJECT> [flags]

clor linear initiative project list

List projects in an initiative

clor linear initiative project list <INITIATIVE> [flags]

clor linear initiative project remove

Remove a project from an initiative

clor linear initiative project remove <INITIATIVE> <PROJECT> [flags]

clor linear initiative show

Show one initiative

clor linear initiative show <INITIATIVE> [flags]

clor linear initiative unarchive

Unarchive an initiative

clor linear initiative unarchive <INITIATIVE> [flags]

clor linear initiative update

Update an initiative with explicit clear operations

clor linear initiative update <INITIATIVE> [flags]

clor linear issue

Read and change Linear issues, comments, and relations

Use when: - the user needs Linear issue records

clor linear issue [flags]

clor linear issue archive

Archive an issue

clor linear issue archive <ISSUE> [flags]

clor linear issue comment

Read and change issue comments

Use when: - the user needs Linear comment records

clor linear issue comment [flags]

clor linear issue comment create

Create an issue comment in Markdown

clor linear issue comment create <ISSUE> [BODY] [flags]

clor linear issue comment delete

Delete an issue comment

clor linear issue comment delete <COMMENT> [flags]

clor linear issue comment list

List comments on an issue

clor linear issue comment list <ISSUE> [flags]

clor linear issue comment resolve

Resolve a comment thread

clor linear issue comment resolve <COMMENT> [flags]

clor linear issue comment unresolve

Unresolve a comment thread

clor linear issue comment unresolve <COMMENT> [flags]

clor linear issue comment update

Update an issue comment in Markdown

clor linear issue comment update <COMMENT> [BODY] [flags]

clor linear issue create

Create an issue with typed workspace references

clor linear issue create <TITLE> [flags]

clor linear issue delete

Move an issue to recoverable trash

clor linear issue delete <ISSUE> [flags]

clor linear issue list

List issues with exact workspace filters

clor linear issue list [flags]

clor linear issue relation

List, add, and remove issue relations

Use when: - the user needs Linear relation records

clor linear issue relation [flags]

clor linear issue relation add

Add a typed relation from one issue to another

clor linear issue relation add <ISSUE> <RELATED_ISSUE> [flags]

clor linear issue relation list

List relations sourced from an issue

clor linear issue relation list <ISSUE> [flags]

clor linear issue relation remove

Remove an issue relation

clor linear issue relation remove <RELATION> [flags]

clor linear issue show

Show one issue

clor linear issue show <ISSUE> [flags]

clor linear issue unarchive

Unarchive an issue

clor linear issue unarchive <ISSUE> [flags]

clor linear issue update

Update an issue with explicit set and clear operations

clor linear issue update <ISSUE> [flags]

clor linear label

Read and change Linear issue labels

clor linear label [flags]

clor linear label create

Create an issue label

clor linear label create <NAME> [flags]

clor linear label delete

Delete an issue label

clor linear label delete <LABEL> [flags]

clor linear label list

List labels

clor linear label list [flags]

clor linear label show

Show one label

clor linear label show <LABEL> [flags]

clor linear label update

Update an issue label

clor linear label update <LABEL> [flags]

clor linear project

Read and change Linear projects, progress updates, and milestones

Use when: - the user needs Linear project records

clor linear project [flags]

clor linear project create

Create a project across one or more teams

clor linear project create <NAME> [flags]

clor linear project delete

Move a project to recoverable trash

clor linear project delete <PROJECT> [flags]

clor linear project list

List projects

clor linear project list [flags]

clor linear project milestone

Read and change project milestones

Use when: - the user needs Linear milestone records

clor linear project milestone [flags]