From autoloop
Use when running a development loop that should report its status (project, phases, commits) to an Autoloop dashboard. Reports via the bundled autoloop CLI as the loop progresses. Requires AUTOLOOP_API_KEY in the environment and a one-time `autoloop init`.
How this skill is triggered — by the user, by Claude, or both
Slash command
/autoloop:autoloop-reportingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Report the loop's status to Autoloop as you work, using the bundled **`autoloop`**
Report the loop's status to Autoloop as you work, using the bundled autoloop
CLI (this plugin adds it to your PATH; it needs Node 22+, no other deps).
Reporting is best-effort observability — it must never block or derail the
actual development work.
AUTOLOOP_API_KEY must be set in the environment — a per-user key minted in
the Autoloop app under API keys (you must be a member of the team you report
to). The CLI never reads the key from a file.
Run autoloop init once in the loop's working directory to write .autoloop.json:
autoloop init --team <teamId> --project <slug>
(The CLI defaults to the hosted Autoloop API; pass --url <apiUrl> only to point
at a different deployment.)
| Loop moment | Command |
|---|---|
| Project start (once) | autoloop init … then autoloop project set --title "<title>" --status running --design-file <plan-or-spec> |
| Entering a phase | autoloop phase start <phaseId> --name "<name>" --order <n> |
| After each git commit | autoloop commit |
| Leaving a phase | autoloop phase set <phaseId> --status completed (or failed) |
| Loop end | autoloop project set --status completed (or failed / cancelled) |
autoloop commit reads the latest git commit (sha, message, author, time) and
attaches it to the current phase — just run it right after you commit.phaseId is yours to choose (e.g. build, design); phase set reuses the
name/order you gave at phase start.autoloop command prints a warning (bad key, not a team
member, network blip), note it and keep going — never abort the loop over a
reporting failure. By default such failures exit 0; only pass --strict if you
deliberately want reporting failures to be fatal.status must be one of
queued | running | blocked | paused | completed | failed | cancelled. IDs
(teamId, slug, phaseId) must match ^[a-z0-9._-]+$ (lowercase, no spaces or
slashes).project set (with title) before phases, and phase start
before the phase set/commit for that phase. The CLI is resilient if you don't,
but this keeps the dashboard accurate.autoloop init --team acme --project web
autoloop project set --title "Acme Web" --status running --design-file docs/plan.md
autoloop phase start build --name "Build" --order 1
# … do work, git commit …
autoloop commit
autoloop phase set build --status completed
autoloop project set --status completed
If
autoloopisn't found on yourPATH, invoke it directly:node "${CLAUDE_PLUGIN_ROOT}/bin/autoloop" …
See CODEX.md in this directory for the same commands framed for a Codex-driven loop.
Provides CDSS development patterns for drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), and alert classification integrated into EMR workflows.
npx claudepluginhub openloopagentics/autoloop --plugin autoloop