Content
82%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exemplary lean workflow: numbered steps, quantified thresholds, a concrete output template, modification guardrails, and a worked example, with the heavy case data correctly offloaded to a real bundle reference. The main fixes are small operational gaps: specify the `self-testing.mdc` path, define first-run behavior when no retrospective marker exists, and add a post-change validation step.
Suggestions
Define first-run behavior when no `<!-- Retrospective completed -->` marker exists (the current case-studies.md contains zero markers) — e.g. 'If no marker exists, analyze the last 10 cases'.
Give the full path for `self-testing.mdc` and either create `references/patterns.md` or note it should be created on first use, since the bundle currently lacks it.
Add a post-application validation step: after Step 5, re-check that `self-testing.mdc` is still under 150 lines and retains the 9-section structure before appending the completion marker.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and instruction-only — no concept explanations, no padding; every section is a rule or a template (e.g. "Max 150 lines — if exceeded, consolidate similar items", the compact report block, and the 3-line worked example). This matches 'Lean and efficient; assumes Claude's competence; every token earns its place'. | 5 / 5 |
Actionability | Guidance is largely executable: exact thresholds ("fewer than 3 new cases", "2+ times marked NEW", "3+ times already covered"), a fill-in report template, and four concrete modification rules. Minor gaps keep it below 5: no path for `self-testing.mdc`, and no specification of first-run behavior when no `<!-- Retrospective completed -->` marker exists (the shipped case file contains none). | 4 / 5 |
Workflow Clarity | Six clearly sequenced steps with real checkpoints: an explicit stop gate ("If fewer than 3 new cases, report 'Not enough new cases' and stop") and a user-confirmation gate before modifying `self-testing.mdc`, so the destructive/batch cap does not apply. It falls short of anchor 5 because there is no feedback loop after applying changes (no re-validation that the 150-line limit or 9-section structure still holds) and the unhandled no-marker first run is a gap in the sequence. | 4 / 5 |
Progressive Disclosure | The ~48-line body stays an overview and pushes the bulk (671-line case data) into `references/case-studies.md`, which exists in the bundle and is one level deep. Remaining gaps fit anchor 4: `references/patterns.md` is named as an overflow destination but does not exist in the bundle, and the reference is cited by the long path `.claude/skills/1k-retrospective/references/case-studies.md` rather than a clean relative link, slightly blurring navigation. | 4 / 5 |
Total | 17 / 20 Passed |