Linearis

linearis

Drive Linear.app from the shell via the linearis CLI (JSON-only output; linear is an alias). Do not guess the command surface — the CLI documents itself, and this skill teaches the protocol, not the flags.

Preflight (reactive — branch on the CLI's own output; don't pre-run checks every turn)

• Not installed — if the shell reports command-not-found, tell the user linearis isn't installed and offer npm install -g linearis. As a no-install fallback, prefix commands with npx linearis@latest (adds cold-start latency and needs network per call — fallback, not default). Never silently npm install -g.
• Auth required — any command may fail with this envelope on stderr and exit code 42: { "error": "AUTHENTICATION_REQUIRED", "action": "USER_ACTION_REQUIRED", "instruction": "Run 'linearis auth login' …", "exit_code": 42 }. Detect it by exit_code === 42 / error === "AUTHENTICATION_REQUIRED" (not paraphrased text) and surface the CLI's own instruction. linearis auth login is an interactive browser flow you cannot complete — hand it to the user.
• Invalid invocation — an unknown command or option, a wrong argument count, or a command group named without a subcommand fails on stderr with exit code 2: { "error": "UNKNOWN_COMMAND", "message": "…", "suggestion": "Did you mean read?", "command": "linearis issues", "available_commands": [...], "instruction": "Run 'linearis issues usage' …", "exit_code": 2 }. Recover from the envelope, not by guessing: pick from available_commands, or run the instruction. error is one of UNKNOWN_COMMAND, UNKNOWN_OPTION, MISSING_ARGUMENT, MISSING_REQUIRED_OPTION, MISSING_OPTION_ARGUMENT, TOO_MANY_ARGUMENTS, MISSING_SUBCOMMAND, INVALID_USAGE. A bare group (linearis issues) is a failure, not a request for help.
• Updates (advisory, never blocking) — optionally run linearis version check once → { current, latest, channel, updateAvailable }. If updateAvailable is true, mention it and ask the user before npm install -g linearis@latest, honoring channel (don't move a next user to latest). npm can hang or rate-limit; on any timeout/error just proceed with the installed version. Read the plain installed version with linearis version (JSON), not --version.

Discover, then act

1. Run linearis usage once for the list of domains (issues, projects, cycles, …).
2. Run linearis <domain> usage for a domain's full command and flag reference before acting.
3. Never invent flags or subcommands — usage is authoritative and always current.

Output

Every command prints JSON on stdout. Shape it at the source with the global --fields identifier,title,state.name and --compact — no external binary, works on Windows and fresh containers. Reach for jq only for complex reshaping, and fall back to raw JSON if jq is absent.

Invariants worth knowing (everything else lives in usage)

• IDs are forgiving: pass a UUID, team key (ENG), issue identifier (ABC-123), or name interchangeably. Reference tickets by identifier.
• issues create requires --team; some filters need a scope flag — confirm in usage rather than memorizing.
• Threaded discussion lives under issues discuss / discussions / replies / reply. The top-level comments domain is a deprecated facade (still works) — prefer the issues discussion commands. Record non-trivial progress in a discussion thread and keep the description in sync on status changes.
• files download <url> only fetches Linear storage URLs (uploads.linear.app); files upload returns an assetUrl you can embed; issues read --with-attachments lists linked resources (PRs, docs, URLs) — references, not necessarily downloadable files.

For anything not covered here, linearis <domain> usage is the reference.