Content
50%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 lean, sectioned body with one executable example and a sequenced procedure, but it is held back by generic boilerplate sections, a single code example for many rules, unspecified validation commands, a broken reference path, and an off-topic stray bullet ("Redact secrets, tokens, credentials, and sensitive data by default.") misplaced under Type Safety.
Suggestions
Fix the reference path from `Infrastructure/references/deep-guidance.md` to `references/deep-guidance.md` and either list or remove the unmentioned bundle files (contract.yaml, evals.yaml, task-profile.json).
Replace "Run the relevant local checks" with a concrete command (e.g., `npx tsc --noEmit` on touched files) so the Validation feedback loop is executable rather than implicit.
Cut the generic Philosophy/Procedure/Anti-patterns boilerplate and the stray "Redact secrets" bullet, and add 1-2 more executable examples covering the remaining rules (e.g., a discriminated union or an IO-boundary guard).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Bullets are terse and no basic concepts are over-explained, but the generic Philosophy/Procedure/Anti-patterns boilerplate ("Optimize for clear, verifiable outcomes with the minimum necessary changes"), the redundant Table of Contents, and a "When to use" section duplicating the frontmatter description are unnecessary padding. Mostly efficient with several sections that could be trimmed — anchor 3, not anchor 2 since most lines do earn their place. | 3 / 5 |
Actionability | One complete executable example exists (the `parsePort` function) plus concrete per-rule directives ("Avoid `any`", "Prefer named exports over default exports"), but only one code example covers roughly nine rules, and validation guidance names no command ("Run the relevant local checks"). The gaps exceed anchor 4's "minor gaps"; anchor 3's "some concrete guidance but incomplete" fits. | 3 / 5 |
Workflow Clarity | The Procedure section gives a sequenced three-step flow and Validation specifies fail-fast plus re-run-after-fix behavior, so a feedback loop exists. However, checkpoints are implicit — no concrete check command (e.g., `tsc --noEmit`) is ever named, matching anchor 3's "checkpoints missing or implicit" rather than anchor 4. | 3 / 5 |
Progressive Disclosure | Sections are organized and the deep-guidance reference is well-signaled with a "Read when:" clause, but the path `Infrastructure/references/deep-guidance.md` does not match the actual bundle file `references/deep-guidance.md`, and three of the four files in `references/` (contract.yaml, evals.yaml, task-profile.json) are never mentioned in SKILL.md. Scored against the actual bundle structure, this is "some structure but could be better organized" — anchor 3. | 3 / 5 |
Total | 12 / 20 Passed |