Skip to content

Releases: YoungApple/growth-retrospective-skill

Release list

v0.2.0 — Forcing function (anti-busywork)

v0.2.0 — Forcing function (anti-busywork) Latest
Latest

Choose a tag to compare

View all tags
@YoungApple YoungApple released this 25 May 08:37
v0.2.0
cf6fcbf

What's new in v0.2.0

The big change is Step 0 — Action Audit. The skill now refuses to run a new scan if the last retro's level-up actions are untouched.

Why

Persona review (returning-user simulation) surfaced the #1 stickiness defect: a second-time run produced near-identical output because no action was completed between runs. The skill felt like busywork.

v0.2.0 turns it into an accountability loop: items don't pile up because new scans don't run until old commitments close (or get explicitly killed).

How it works

Before any new scan, the skill:

  1. Reads the current docs/learning-domains.md
  2. Extracts every `How to level up` action's named artifact (`STATS-CHEATSHEET.md`, `gemini:probe` script, `docs/entity-erd.md`, etc.)
  3. Checks the filesystem — did the artifact get built?
  4. If ≥3 actions are pending >14 days AND 0 completed since last run, it pushes back with three options: Commit (pick one to make active), Deprioritize (kill items explicitly), or Override (proceed but log why).

Other changes

  • CONTRIBUTING.md — high-value contribution targets ranked. The most-wanted contribution is a real worked example.
  • .github/ISSUE_TEMPLATE/ — three templates: bug / worked-example / new-agent-support. The worked-example one surfaces L3 outcome data (graduations + stickiness) — the only data point that validates the "shrinking gap" claim.
  • .github/PULL_REQUEST_TEMPLATE.md — checklist enforces cross-agent testing + line-count discipline.

Validation

Output-layer quality (L1 in EVAL-FRAMEWORK.md) was validated via iteration-1 → iteration-2 evals. All v1 defects fixed in v2 (language compliance, 5-category sweep, demotion check). v2.5 (this release) builds on that with the forcing function — but the forcing function itself has not yet been validated against a third iteration of evals.

Honest TODOs

  • L2 (acceptance) and L3 (outcome) data: zero. The skill needs real-user data to verify the shrinking-gap claim.
  • The forcing function's heuristics ("≥3 pending + 0 completed") are reasonable defaults but not tuned against real distributions.
  • Cross-agent compatibility tested in Claude Code only. Codex / Antigravity testing welcome via PR.

Install

```bash
git clone https://github.com/YoungApple/growth-retrospective-skill ~/.claude/skills/growth-retrospective
```

See README for Codex / Antigravity install paths.

Assets 2
Loading