pm-recurring-cycle

name: pm-recurring-cycle description: Run deterministic project-manager recurring consultation cycles in project/second-brain. Use only for explicit PM-cycle intents such as "run PM recurring cycle", "consult project manager", "resolve PM consultation", or "advance PM stage/task state" via tools/sb_agent_run_cycle_v0.sh. Do not use for direct feature execution or non-PM workflows.

PM Recurring Cycle

Outcome

Run one PM loop that keeps stage visibility current, asks or resolves exactly one consultation decision, and records all mutations in scene artifacts and the mutation ledger.

Required State Surfaces

• scene/agent/project_manager/status_v0.json
• scene/agent/project_manager/ideas_v0.json
• scene/agent/project_manager/consultation_queue_v0.json
• scene/task_queue/v0.json
• scene/authority/registry_v0.json
• scene/ledger/mutations_v0.jsonl

Trigger Boundaries

• Trigger on explicit PM governance-cycle requests.
• Trigger when the user asks to resolve numbered PM consultation choices.
• Skip when the user asks for implementation work outside PM state orchestration.
• Use references/branch_templates.md for deterministic field-level branch updates.

Cycle Workflow

1. Inspect current state and pending consultation.

jq '.stage_summary, .signals, .next_consultation_due' scene/agent/project_manager/status_v0.json
jq '.items[] | select(.state=="pending")' scene/agent/project_manager/consultation_queue_v0.json
jq '.tasks[] | select(.owner_agent=="agent/project_manager_v0") | {task_id,state,last_progress_at}' scene/task_queue/v0.json

2. Select exactly one branch:

• consult: resolve one pending consultation from an explicit user choice.
• kickoff: if none pending, add one next consultation and update stage focus.
• status-only: refresh stage signals with no new consultation.

3. Apply updates via a small local Python mutator script (write only target files).

4. Execute one deterministic wrapper cycle (never write tracked artifacts directly):

./tools/sb_agent_run_cycle_v0.sh \
  --agent-id agent/project_manager_v0 \
  --targets scene/agent/project_manager/status_v0.json,scene/agent/project_manager/ideas_v0.json,scene/agent/project_manager/consultation_queue_v0.json,scene/task_queue/v0.json \
  --reason "<pm cycle reason>" \
  --task-id task/project-manager-recurring-cycle-v0 \
  --phase <phase_slug> \
  --mutation-type UPDATE \
  --inputs OPEN_QUESTIONS.md,scene/task_queue/v0.json,scene/agent/project_manager/consultation_queue_v0.json \
  --exec-cmd "python3 /tmp/<mutator>.py"

5. Verify and report:

jq '.items[] | select(.state=="pending")' scene/agent/project_manager/consultation_queue_v0.json
jq '.stage_summary, .signals' scene/agent/project_manager/status_v0.json
jq '.tasks[] | select(.task_id=="task/project-manager-recurring-cycle-v0" or .task_id=="task/oq3-autonomy-risk-policy-propose-v0") | {task_id,state,last_progress_at}' scene/task_queue/v0.json
tail -n 5 scene/ledger/mutations_v0.jsonl

Decision Contract

• Keep PM in delegator/planner mode by default.
• Consult the human before high-impact changes.
• Resolve only one consultation per cycle unless user asks for batching.
• Keep mutations scoped to PM scene surfaces and assigned task rows.

Output Contract

• Return:
  • what decision was processed,
  • which files changed,
  • resulting task/stage state,
  • the next explicit choice for the human.