Content
88%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.
A dense, highly actionable workflow skill that front-loads three non-negotiables, sequences an 8-step process with explicit validation and feedback loops, and is rich in genuinely non-obvious repo-specific knowledge (tooling caveats, cost notes, statistical sizing). Its two weaknesses are modest: minor rhetorical padding in the decision-framing prose, and a monolithic ~290-line body whose PostHog-specific pattern catalogs could be split into a reference file.
Suggestions
Move the frontend (Jest/kea/MSW) and backend (pytest) root-cause pattern catalogs out of step 4 into a references/ file (e.g. references/posthog-flake-patterns.md), keeping a symptom-to-cause table inline so the main body stays a navigable overview.
Trim the persuasive framing in step 5 (e.g. "the one case where cost is already proven and value is not") and condense the broken-tests/test-health surface taxonomy into a two-line do/don't — the operational rules already carry the point.
The git bisect usage explanation is close to standard knowledge; keep only the flake-specific caveat (lucky pass per step, or use the CI run-history boundary) and drop the basic start/run commands.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly all content is non-obvious environment-specific fact Claude cannot know (MSW unhandled-request behavior, LISTENER_FINISH_WAIT_TIMEOUT, ci:insights lacking a denominator, the pnpm build:products rebuild cost, the N ≥ 3k sizing math), so it earns its tokens. A few stretches could be trimmed — the rhetorical framing in step 5 ("the one case where cost is already proven and value is not") and the extended broken-tests/test-health taxonomy — keeping it below the every-token-earns-its-place anchor. | 4 / 5 |
Actionability | Fully executable throughout: gh run list/gh api commands with real flags and --json fields, a complete copy-paste bash loop harness judging by exit code, git bisect run with a concrete repro wrapper, the pnpm exec jest variant that skips the rebuild, and a fill-in report template. Placeholders (<test command>, <run-id>) are the only substitutions needed. | 5 / 5 |
Workflow Clarity | Eight numbered steps with an explicit five-level reproduction escalation ladder, a validation checkpoint in step 7 sized from the measured pre-fix rate (N = max(3k, 20)), an explicit feedback loop ("Any failure in the loop → back to step 4"), a three-outcome decision table, and hard safety gates — deletion and quarantine require explicit user approval. Validation is present at every risky juncture. | 5 / 5 |
Progressive Disclosure | Well-sectioned file with clear headers, and external pointers are explicitly signaled with paths and what each contains (docs/internal/ci-things-already-tried.md, /writing-tests, frontend/src/mocks/jest.ts) — all one level deep, none nested. However, it is a ~290-line single file: the PostHog-specific frontend/backend root-cause pattern catalogs (kea/MSW/pytest details) are inline content that could live in a references/ file to slim the main body, which keeps it below the appropriately-split anchor. | 4 / 5 |
Total | 18 / 20 Passed |