Content
77%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 audit procedure with real validation checkpoints and error recovery — commands are copy-paste ready and every rule is repo-specific rather than generic. Its weaknesses are token efficiency (the report-shape rule is enforced through repetition rather than one statement plus a check) and a monolithic single-file layout that inlines detail a references/ bundle would carry better.
Suggestions
Split the tautological-test shapes and the "cost more than they catch" catalogue into a references/ file (e.g. references/test-shapes.md), keeping a one-line pointer in Step 4 — that alone would cut roughly a quarter of the body.
State the five-heading rule once in Step 0 and have the mid-run and final checks reference it ("re-run the Step 0 shape check") instead of restating the heading list and the disallowed `## Summary`/`## Verdict` headings three times.
Move the whole-package sweep mode paragraph out of Step 4 into its own reference or a clearly separate section — it is a different task from the branch audit that dominates the file, and interleaving it makes the branch workflow harder to follow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and repo-specific with no explanation of concepts Claude already knows, but the five-heading shape rule is restated four times (Step 0, the bold "only ever edited, never rewritten" paragraph, the mid-run "Check the shape once" block, and the final "Check the file before you finish" section) and the disallowed-heading list (`## Summary`, `## Verdict`, `## Scope and method`, `## Bottom line`) is repeated three times. Not 4: that repetition exceeds "minor instances that could be trimmed"; not 2: nearly every sentence carries non-generic, actionable instruction. | 3 / 5 |
Actionability | Fully executable throughout: exact commands for base resolution (`git config --get branch.$(git branch --show-current).gh-merge-base`), builds and tests (`bun run --cwd <workspace-path> test:run`), mutation testing (`STRYKER_TESTS='tests/lib/money.test.ts' bun run --cwd cire/api test:mutation --base "$DIFF_BASE"`), a verbatim heredoc report skeleton, grep shape checks with expected outputs ("It must print `5`"), and an exact four-field finding format. Specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | Clear Step 0–5 sequence with explicit validation checkpoints (the shape check "as soon as the first gap is in the file", the final two grep counts both required to print 5) and genuine error-recovery loops: restore the skeleton and re-edit if the count is low, and defined handling for `no changed source lines`, `No tests were executed`, and a `0.00` tests-per-mutant run. "No step is a stop" tells the model how to proceed past every failure mode. | 5 / 5 |
Progressive Disclosure | No bundle files exist and all content sits inline in one ~290-line file. Section structure and navigation are good, but bulk detail that belongs in reference files is inlined: the tautological-test shape catalogue, the mutation-testing log-interpretation rules, and the whole-package sweep mode paragraph. Not 4: the split is more than a "minor organization gap" for a file this size; not 2: the file is well-sectioned, not a monolithic wall. | 3 / 5 |
Total | 16 / 20 Passed |