runs

/capusqa:runs — watch and control runs (Runs)

The dashboard Runs page, in chat. It reads and controls the daemon's live state, so every view matches the dashboard and every action persists. (To launch a test, that's /capusqa:test; for findings, /capusqa:issues.)

Reaching the daemon

BASE = the capusqa URL in .mcp.json (default http://127.0.0.1:7777; honor a custom --port). GET routes need no auth; POST routes add -H "Authorization: Bearer $CAPUSQA_DASHBOARD_TOKEN" only if that env var is set. curl -s $BASE/api/status first — no response ⇒ capusqa serve (see /capusqa:help).

List runs — "what's running / show my runs"

curl -s $BASE/api/runs → {runs:[…]}, newest first (id, app, created_at, status, session counts, finding counts, config). Render a compact table; flag anything still active or with hot (critical/high) findings. The dashboard shows each run's status as Active, Completed, Stopped, or Couldn't run — use the same words.

Drill into a run — "show me run X"

curl -s $BASE/api/runs/<run_id> → {run, personas, sessions, findings, finding_groups, clusters, rules}. Summarize per-session status + persona + verdict, the findings (severity / type / summary), and progress. The dashboard labels sessions Running, Starting, Done, Waiting, or Needs attention — match it. (404 unknown run → list /api/runs, let them pick.)

Live — "what's on screen now / watch it"

curl -s $BASE/api/live → {live:[{session_id, run_id, goal, persona_name, steps_taken, …}]}. Live screenshots stream at $BASE/api/live/<session_id>.png — point them at the dashboard to watch, or fetch a frame. For one session's full step trace use the trace_get(session_id) MCP tool.

Steer / stop a live session

• Steer: curl -s -X POST $BASE/api/sessions/<session_id>/steer -H 'content-type: application/json' -d '{"message":"<note>"}' (authed). The note rides the agent's next tool result (no interruption); recorded as an operator step. 409 → not live.
• Stop one session: curl -s -X POST $BASE/api/sessions/<session_id>/stop (authed). Works on a live session OR a terminal failed (needs-attention) one — either way it becomes stopped, which counts as finished. The run's other sessions keep playing.
• Restart a failed session: curl -s -X POST $BASE/api/sessions/<session_id>/restart (authed). Re-queues a needs-attention session and (re)spawns a player to replay it. 409 → the session isn't failed.
• Stop a whole run: curl -s -X POST $BASE/api/runs/<run_id>/stop (authed) → terminates the runner + aborts that run's live sessions + finishes the run.