Work mesh

Presence and per-turn activity are automatic. Do not call deleted core/scripts/work-mesh.sh / pack listen / watch. Use hq-cli:

Daemon

hq mesh daemon install
hq mesh daemon status
hq mesh daemon doctor

One resident process per machine. Flushes ~/.hq/work-mesh/spool.jsonl, holds unverified-company events, publishes MQTT presence (LaunchAgent ai.getindigo.hq-mesh-daemon / systemd hq-mesh-daemon.service).

Context

hq mesh context reconcile --observation-file <path> --machine
# or: hq mesh context reconcile --observation-json '<json>' [--machine] [--offline]
hq mesh context default get|set|clear
hq mesh context untracked <sessionId>
hq mesh context organize --session <sid> --decision <id> --option <id>
hq mesh context correct --session <sid> --to-company <slug> [--project <id>] [--task <id>]

reconcile takes observation input only (no --session). organize / correct / untracked keep --session (or the sessionId argument).

Create options also require a title: append --create-project '<approved project title>' or --create-task '<approved task title>' to organize. Reuse an already approved title, or obtain it before submitting. Existing choices need neither flag.

Project registration after planning

Local board.json is not the server PROJECT_VIEW. Before registration, persist and verify the canonical server view through the supported authenticated HQ client, using the approved company API and exact project ID. Never print or paste tokens. Complete these steps in order:

1. Read before writing: authenticated GET /v1/work-mesh/projects/{projectId}?companyUid={companyUid}. Verify the returned companyUid and projectId exactly match the approved plan. Only a confirmed 404 means the view is absent; stop on authorization, transport, or other errors. If the existing view already matches the required stories and repos, reuse it without a PUT.
2. Persist the server view: authenticated PUT /v1/work-mesh/projects/{projectId} with companyUid, name, description, complete stories, and complete repos in the JSON body. Map userStories to story IDs and content; map all metadata.repos entries to supported repo records. Do not send the local board entry or an empty/stub view as the plan. For an existing view, merge by story ID and repo identity, preserve live story statuses and other server activity fields, and retain unrelated server entries. Never replace live statuses with local passes: false or default queued values. The current public PUT is a full replacement and has no conditional-version guard: do not assume sending expectedVersion protects it. Coordinate exclusive writing before changing an existing view; if that cannot be established, stop and report the required merge instead of risking concurrent status updates.
3. Verify persistence: repeat the authenticated GET /v1/work-mesh/projects/{projectId}?companyUid={companyUid} after PUT (or inspect the successful GET when reusing an unchanged view). Verify the exact company and project IDs, every planned story ID and its content, and every planned repo identity/path. Compare full ID sets and mappings, not just counts; confirm previously live story statuses and retained entries are unchanged. On any missing entry or mismatch, report the view as incomplete and stop.
4. Register only after verification: authenticated POST /v1/work-mesh/projects/{projectId}/register with JSON body {"companyUid":"<resolved company UID>"}. Reuse the same canonical project ID.

This operation requires the existing tenant-scoped project view, creates or reuses its deterministic genesis, and ensures its project channel. Require a successful response containing both threadId and channelId before reporting registration complete. Confirm the channel is readable with the current identity. A Board view, cached thread ID, session presence, or successful note alone is not proof of registration or channel creation.

If registration fails or the deployed server does not support this operation, report registration/channel creation as incomplete and preserve the plan and canonical ID. Inspect partial state before recovery; never create a replacement project or weaken tenant ownership checks. Presence hooks remain automatic and do not perform this project-registration step.

Session (manual only)

hq mesh session task-status --session-id <sid> --enqueue --seq <n> --task-id <id> --status queued|in_progress|review|done
hq mesh session blocked --session-id <sid> --enqueue --seq <n> --reason "<short>"
hq mesh session note --session-id <sid> --enqueue --seq <n> --summary "<<=280 chars>"
hq mesh session flush

Session verbs use --session-id (not --session). Hooks already enqueue session_start / turn_start / turn_end / session_end. Policy: core/policies/hq-project-work-mesh-reporting.md.