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.
The body is highly actionable and the gated workflow is clearly sequenced with concrete thresholds and recovery paths, but it is self-inconsistent on conciseness and progressive disclosure: it repeats its core points multiple times and inlines two large reference tables that its own routing rules would send to separate files. Extracting those tables into a references/ file and de-duplicating the size-is-not-the-trigger messaging would tighten it considerably.
Suggestions
Extract the 11-row "Rationalizations to catch in yourself" table and the 10-row "Anti-Patterns to Catch" table into a references/ file (e.g. references/rationalizations.md), keeping a 3-4 row summary inline — this follows the skill's own rule that tables with 10+ rows route to a separate doc.
De-duplicate the "size is not the trigger" messaging: it is stated in the intro, the STOP paragraph, and at least three rationalization rows; state it once prominently and cut the redundant rows (e.g. merge "Guardian would obviously say yes" and "I've mentally walked through this already", which make the same point).
Trim the carve-out section: the flag-file mechanization detail (~26 lines) repeats the approval conditions twice (bullets and prose); a single tight list of the four narrow conditions plus one line on the flag mechanism would preserve the safety net at half the length.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~250-line body contains only non-obvious, project-specific policy (no explaining of things Claude already knows), but key points are repeated: "regardless of edit size" / "size is not the trigger" appears at least four times, and rationalization rows like "Guardian would obviously say yes" and "I've mentally walked through this already" overlap heavily with each other and with the spirit-vs-letter opening. | 3 / 5 |
Actionability | Fully executable for an instruction-only skill: the six-question pre-write checklist has concrete thresholds (200-line limits, 30-line @-import rule), a yes/no litmus test with worked examples, a routing flowchart, per-destination format rules, a copy-ready pushback script, and BAD/GOOD condensation examples. | 5 / 5 |
Workflow Clarity | A clear ordered sequence (six numbered questions) with explicit checkpoints and feedback loops: over-threshold files trigger "extract or condense first" recovery, the pushback path handles user insistence, and the carve-out deviation path specifies disarm/re-arm steps. This matches the checklist-with-error-recovery anchor. | 5 / 5 |
Progressive Disclosure | Section headers are clear and navigation within the file is easy, but it is a single monolithic body with no bundle files: the 11-row rationalizations table and 10-row anti-patterns table are reference material inlined in SKILL.md — the skill's own anti-patterns table says tables with 10+ rows should be routed to a separate doc. Sits between the inlined-content (2) and good-structure (4) anchors. | 3 / 5 |
Total | 16 / 20 Passed |