git:20260528.a4e7dfe to git:20260729.2975ca8

6 added, 0 removed. Audit A to A.

---

name: kill-or-keep

description: Quarterly portfolio review with forced verdicts (DOUBLE-DOWN / MAINTAIN / MINIMAL / KILL) for every product.

---

# /kill-or-keep

Usage: `/kill-or-keep` (run quarterly, end of quarter)

## What this does

Invokes `agents/portfolio/kill-or-double-down/`. Produces a full portfolio review document at `brain/decision-log/portfolio-review-YYYY-QN.md`.

## What you do as the assistant

1. Confirm with user: "this is a quarterly review and will force verdicts. Proceed?"

2. Read all products' `product.config.json`, `metrics.md`, last 90 days of `decisions/`.

3. Pull current metrics via telemetry-aggregator.

4. Ask user: "estimated hours per week on each product over last quarter" — fill in.

5. Generate the review per the agent's output spec.

6. For each KILL verdict, generate the sunset plan in full.

7. Save to file. Print summary in chat.

+ 7b. **Factory curator pass** (rides here quarterly — v2.20 MECE curator, now measured):

+ run `bun run skill-report --days 90`. `dormant` skills are archive candidates

+ (`factory/skills/_archive/` — archive, never delete), `demoted` skills get a

+ look before anything trusts them, and guard-sunset questions cite

+ `bun run defect report` counts. Record pin / consolidate / archive per the

+ skill-authoring playbook's curator rules.

8. Update each killed product's `product.config.json` → `status: "sunset_planned"` + `sunset_date: <T+60>`.

9. Add an entry to `MEMORY.md` if portfolio-level patterns emerged.

## Warning

This skill produces uncomfortable decisions. Don't soften the verdicts. The whole point is the forcing function.