Content
93%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 tight, command-driven skill: the responsibility split, safety rules (dry-run, parity check, ask before taste decisions), and a validated six-step sequence are all concrete and executable. The only gap is the absence of explicit error-recovery guidance when a checkpoint (dry-run, status check, first CI run) surfaces a problem.
Suggestions
Add a short failure-handling note to the Sequence (e.g., 'If the dry-run plan looks wrong or fallow agent status reports a missing registration, fix and re-run the dry-run before proceeding'), turning the existing checkpoints into validate -> fix -> retry loops.
In step 5, name the concrete baseline command or flag (or point explicitly to the Baselines section of ci-gate.md) so the baseline action is as copy-paste ready as the other steps.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean with zero padding — no concept explanations (e.g., what a package manager or CI provider is), just terse rules ('Resolve every flag from fallow --help. Do not use flags from memory.') and a numbered sequence. Every line instructs; it fully assumes Claude's competence. | 5 / 5 |
Actionability | Concrete, copy-paste-ready commands appear at every step: 'fallow recommend --format json --quiet', 'pnpm add -D fallow', 'fallow agent install --dry-run --format json --quiet', 'fallow agent status --format json --quiet', 'fallow audit', plus package-manager variants ('npx fallow recommend', 'pnpm dlx fallow recommend', 'pnpm exec fallow'). Specific tools are named (Oxfmt, Oxlint, Knip, jscpd, dependency-cruiser) and decisions are triaged into auto/default/taste. | 5 / 5 |
Workflow Clarity | The six-step sequence has explicit checkpoints — dry-run before mutating installs ('Run fallow agent install --dry-run ... show the plan, then run fallow agent install'), post-run verification ('Check the result with fallow agent status'), and a baseline safeguard ('save a baseline of the existing debt so that the gate fails only on new findings'). It falls short of the top anchor because there is no explicit error-recovery loop (no guidance on what to do when a step fails or the dry-run reveals a problem). | 4 / 5 |
Progressive Disclosure | SKILL.md is a clear overview that keeps the sequence inline while deferring detail to three real, one-level-deep references, each linked at point of use and described in a References section ('the files to inspect, the responsibility split, and the parity check'). The deep-linked anchor (tooling-detection.md#parity-check) resolves to a real '## Parity check' section, and no content that belongs in a reference is inlined. | 5 / 5 |
Total | 19 / 20 Passed |