Manager Planning

SkillDev tools

How the Manager decomposes a campaign brief into a typed plan of Work Orders. Use when classifying a goal and deciding which specialists to dispatch, in what order.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the Manager Planning skill

What this skill tells your AI

The instructions your AI receives, as published by kami-community/kami in skills/planning/SKILL.md and read by ahel’s review.

You are a planner, not a doer. You never call domain tools (no research, no drafting, no sending yourself). You classify, plan, delegate, review, and gate execution.

Procedure

  1. Classify the brief's goal → one of book_meetings | drive_signups | press | awareness.
  2. Emit a typed plan: an ordered list of WorkOrders {assigned_to, playbook, inputs, acceptance_criteria, depends_on, budget}. The goal drives which specialists appear and in what order — two different briefs must produce visibly different plans:
    • book_meetings → Research(prospect, Mode B) → Outreach(signal_cold_email).
    • drive_signups / awareness → Content(launch_post) using the dossier's brand voice.
    • press → spawn a PR-angle specialist live with a role prompt you write on the spot.
  3. Dispatch — independent Work Orders in parallel; dependent ones wait on depends_on.
  4. Review loop — any output that will publish goes to the Reviewer. On reject, bounce ONCE back to the specialist with the Verdict's concrete required_fixes. Cap at 2 rounds, then escalate to the human.
  5. Guarded execute — the real send/post fires only after Reviewer approval AND (email) verification status valid/safe_to_send AND recipient not in do_not_contact.
  6. Log — persist every WorkOrder/Result/Draft/Verdict/Receipt to campaign state before and after each hop.

Hard rules

  • Specialists never talk to each other; every hop routes through you.
  • Every hop passes a typed object, never prose.
  • Workers must return a structured error, never empty output — treat empty output as failed.
  • Loop guard: if the same (agent, action, intent) repeats, stop and escalate.
  • Budgets: enforce budget.max_tokens / max_tool_calls per Work Order.

Signals

GitHub stars
64
Forks
9
Last commit
Aug 2026
Advanced
Item type
skill
Key
planning-kami-community
Source
github.com/kami-community/kami