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 exceptionally well-structured process skill: every phase has a concrete script gate, artifacts have literal formats, and detail is cleanly deferred to existing one-level references. The only cost of its extreme compression is long multi-clause rules and slight overlap between the Rules section and the summary tables.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is densely packed with operational content and explains nothing Claude already knows — e.g. rule 6 specifies proof formats in a single sentence, and "Run a script rather than read it: its messages say what to change" wastes no tokens. Not 5 because several rules run very long with stacked clauses and there is mild redundancy between the Rules section and the "Who checks what" table (rule 5 vs. the Verifier row, rule 9 vs. the screens row). | 4 / 5 |
Actionability | Fully executable guidance: exact per-gate commands (`validate_checks.py <feature> --execute`, `check_commit.py --message "<msg>"`), a literal invocation convention (`python3 <skill-dir>/scripts/<name>.py` with `--root` and Windows fallbacks), and copy-paste-ready artifact formats (`Run <name>: <command>`, `Proof: run <name> | <line>`, `**Approved**: <date> - user: "<their words>"`, `<capture>.running.png`). Specific examples cover the common cases, matching the top anchor. | 5 / 5 |
Workflow Clarity | The PLAN → CHECKS → BUILD → VERIFY sequence is diagrammed, tabulated per phase with its reference, and gated by scripts at every transition ("Before presenting the plan", "Before starting to build", "Before each commit", "Before declaring done"). Validation is explicit and machine-checked ("validate_verification.py exits 0", "A non-zero exit means stop and fix") with error-recovery feedback loops ("a red gate is said first", findings "closed by a commit"), matching the top anchor. | 5 / 5 |
Progressive Disclosure | The body is a genuine overview: rules, tables, and an artifact tree, with per-phase details split into one-level-deep references (references/plan.md, checks.md, build.md, verify.md, verify-screens.md, memory.md) that all exist and point back to SKILL.md rather than nesting further. Scripts are likewise referenced by purpose, not inlined. Navigation is easy via the Phases and Scripts tables. | 5 / 5 |
Total | 19 / 20 Passed |