create-proposal

---

name: create-proposal description: Use when an out-of-scope item, Challenger-identified failure mode, or NOT-in-Scope row warrants its own future plan (Rule 12). Trigger phrases: "create proposal", "PROP for this", "open a proposal", "log this for later", "this is out of scope but worth tracking". Produces a PROP-NNN file per the plan-proposal template.

Create Plan Proposal (PROP-NNN)

A Plan Proposal is the cheapest possible structured form a concern can take. It is not a plan — it is a backlog entry that the AI writes and the human triages via /promote, /defer, or /reject. Without proposals, every out-of-scope item disappears the moment a plan is approved (Rule 12).

When to trigger

• A Gatekeeper subagent (G1/G2/G3) flagged a candidate while reviewing a plan
• A Challenger output identified a failure mode worth its own future plan
• A NOT in Scope row in a current plan describes work the team will need eventually
• User explicitly says: "open a proposal for X", "log PROP for this", "track this for later"

Prerequisite

A source plan exists (CTX / EXP / PRD or a §-numbered chapter). Proposals must reference what produced them — generic proposals with no source are an anti-pattern.

Steps

1. Read the template: templates/plan-proposal-template.md.
2. Identify the source — which plan, Challenger output, or NOT-in-Scope row triggered this proposal? Capture the id or §-number for the source-plan: field.
3. Determine the proposals folder from the project's CLAUDE.md — typical: context/plans/proposals/.
4. Assign the next PROP number by scanning existing proposals across proposals/, proposals/promoted/, and proposals/rejected/. Never reuse an id, even from rejected proposals — the audit trail must hold.
5. Draft the proposal:
   • Trigger — quote the source NOT-in-Scope row, Challenger failure mode, or Gatekeeper finding
   • Suggested Goal — one sentence; the seed for the eventual plan's ## Goal
   • Why now / why later — escalation vs. backlog, plus the condition under which "later" becomes "now"
   • Suggested Kill Criterion — when does this proposal become obsolete? (mandatory — proposals without exit ramps are anti-pattern)
   • Estimated effort — Tiny / Small / Medium / Large (gut check, not a commitment)
   • Notes — context the human will need when triaging
6. Set frontmatter:
   • proposed-by: ai (always — humans write plans, not proposals)
   • source-gate: G1 | G2 | G3 if a Gatekeeper raised it; omit if human-requested
   • suggested-type: ctx | exp | prd
   • status: pending-human-review
   • Never write decided-by: — that is human-only.
7. Confirm path, id, and content with the user before writing.

Rules

• Proposals are AI-proposed but human-decided. Never /promote a proposal autonomously — that is a human action.
• Cap of 5 proposals per Gatekeeper gate-check. If more candidates exist, list the top 5 and roll the rest into a single Sammel-Notiz paragraph in the verdict report.
• Generic proposals ("we should clean up code") are anti-pattern — too vague to triage. Prefer a sharper proposal or none at all.
• Rejected proposals may be re-opened only by an explicit human action; the AI does not resurrect them.
• After 90 days pending-human-review, the G1 review process auto-defers a proposal — note this in the proposal's notes if it is borderline urgent.

Output

A PROP-NNN_<slug>.md file in context/plans/proposals/, status pending-human-review. The Gatekeeper / governance log records its creation; the human decides next via /promote, /defer, or /reject.