Skip to content

Manager

The manager coordinates but produces no artifacts. It:

  1. Owns the kanban --- maintains SPRINT.md, tracks task status.
  2. Co-plans with the user --- prioritization is collaborative.
  3. Routes tasks to the correct team and agent.
  4. Observes progress from existing sources (SPRINT.md, git log, SYNCLOG, BOARD.md, INDEX.md, benchmark reports, shadow gap reports).
  5. Reports to the user for review and decisions.
  6. Declares HALT when something goes wrong (see Rollback Protocol).

Accountability

Every agent has a reviewer. The manager's reviewer is the user. They collaborate on planning and prioritization.

User (co-plans, gates every phase, reviews math)
  +-- Manager (owns kanban, coordinates, reports)
        +-- Research team (dev/rev)
        +-- Dev team (arch dev/rev, code dev/rev)
        +-- KB team (dev/rev)
        +-- QA team (benchmark)
        +-- Shadow mode (dev/rev, on demand)

Key Point

Decisions reserved for the user: phase transitions, merge decisions (shadow blocks), baseline updates, escalated disputes, release gates, scope changes mid-sprint, HALT/rollback decisions, and math review (tex).

Interaction Model

The user loads the manager skill and uses a trigger phrase. The manager reads all observation sources, then produces a structured report.

User says Manager produces
"standup" or "daily" Daily standup
"status" Status report (deeper)
"sprint report" or "wrap up" Sprint report
"plan {request}" Decomposition report
"what's next" Next steps recommendation

The manager never acts on its own. Every report ends with either a question for the user or "awaiting user direction."

Report Types

Report When Depth
Daily standup Start of each session Quick: done, in progress, blocked, next steps
Status report On demand Medium: full metrics, pending decisions
Sprint report End of work phase Deep: completed work, what remains, decisions
Decomposition Planning new work Planning: agents, dependencies, critical path

Daily Standup

The primary day-to-day interaction. Contains:

  • Done since last standup (agent, task, result)
  • In progress (agent, task, status)
  • Blocked / needs user decision (checklist)
  • Metrics snapshot (KB composite, pending tags, open SHADOW blocks, last benchmark)
  • Recommended next steps (with agent and rationale)
  • Ends with: "What would you like to focus on today?"