Content
92%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.
An unusually actionable, well-sequenced skill: the two-question gate with model answers, the five no's, the cost ladder, and the per-language rules are all concrete and executable, and the bundle correctly offloads the bug catalog and ratchet remediation to two clearly signaled reference files. The only weakness is localized verbosity — the DRF-validation bullet and the time-bomb/frozen-clock passages explain mechanics or narrative that could be trimmed.
Suggestions
Trim the DRF-validation bullet to the repo-specific rules (use SimpleTestCase for serializer validation, keep one DB-backed endpoint test as a wiring guard, let exceptions-hog render the envelope once) and drop the explanation of standard DRF validator mechanics and to_internal_value ordering that Claude already knows.
Tighten the time-bomb and frozen-clock bullets: keep the rules and the pin-the-date-is-not-pin-the-clock distinction, but cut narrative asides like "that misattribution, not the fix, is where the time goes" and the incident-style framing of why TTL drift reads as a PR-caused failure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense, imperative, repo-specific rules with just enough rationale to apply them to novel cases ("Never commit `.only`", "A test class only takes a database it uses"), but the DRF bullet explains mechanics Claude already knows ("DRF field validators (`required`, type coercion, `choices`, `min/max`, regex) and `validate_<field>` methods run inside `Serializer(data=...).is_valid()`") and the time-bomb section carries trimmable narrative ("that misattribution, not the fix, is where the time goes"). It is not 3 because the over-explanation is localized to a couple of passages rather than pervasive padding. | 4 / 5 |
Actionability | Fully executable guidance throughout: a complete `SimpleTestCase` code example, runnable commands ("grep -rlE '^\s*TTL ' posthog/models/ posthog/clickhouse/ --include=*.py"), named tools and helpers ("self.captureOnCommitCallbacks(execute=True)", "create_person", "freeze_time", "jest.setSystemTime()"), and model answers for the gate questions ("if someone makes `parse_filters` drop the `team_id` clause, this fails"). It is not 4 because the common Python and Jest cases are covered with copy-paste-ready specifics rather than having minor gaps. | 5 / 5 |
Workflow Clarity | The decision procedure is clearly sequenced — two-question gate, then the five no's, then pyramid and per-language rules, then the PR justification line — with explicit validation checkpoints and failure paths ("If you can't answer it concretely — **do not write the test**"; "If you can't write that line, you've found a test that shouldn't be in the PR"). It is not 4 because every stage has an explicit pass/fail check rather than most; this is a decision gate, not a destructive or batch operation, so no validation cap applies. | 5 / 5 |
Progressive Disclosure | The operational gate stays inline while deep material is split into well-signaled, one-level-deep references — "[references/mistakes-we-make.md](references/mistakes-we-make.md)" ("a catalog of the bug shapes we actually ship") and "[references/database-free-test-classes.md](references/database-free-test-classes.md)" — both real bundle files whose contents match the body's descriptions, with no nested references. It is not 4 because the split is clean and every reference is announced with what it contains. | 5 / 5 |
Total | 19 / 20 Passed |