Content
38%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 well-organized with genuinely useful concrete material (hooks config, CLI commands, scope-decision tables), but it is overlong and padded with version history and rationale, and — most seriously — it is shipped as a single file whose referenced scripts and config files are missing from the bundle, breaking both executability and navigation.
Suggestions
Ship the referenced files (`hooks/observe.sh`, `hooks/hooks.json`, `instinct-cli.py`, `config.json`) in the bundle, or remove/correct the paths so every reference resolves to a real file.
Move the version-comparison tables (v2.1 vs v2.0, v2 vs v1), backward-compatibility notes, and privacy section into a one-level-deep reference file (e.g. CHANGELOG.md or MIGRATION.md) and keep SKILL.md as a lean overview.
Add validation checkpoints to the Quick Start workflow — e.g. after enabling hooks, verify `observations.jsonl` is growing; before/after promote, confirm instinct counts with `/instinct-status` or `--dry-run`.
Remove the duplicated command listing by keeping either the Quick Start block or the 命令 table, not both.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~350-line body is noticeably padded: two version-comparison tables (v2.1 vs v2.0 and v2 vs v1), a "为什么用钩子而非技能" rationale section, backward-compatibility prose, and a privacy section largely restate what Claude can infer, and the command list appears twice (Quick Start §3 and the 命令 table). Time-sensitive version history is not isolated in a deprecated/old-patterns section. | 2 / 5 |
Actionability | There is concrete, mostly executable guidance (exact hooks JSON for settings.json, mkdir commands, a config.json example with a key table, and `python3 instinct-cli.py promote [--dry-run]` CLI calls). However, the referenced `observe.sh`, `hooks/hooks.json`, `instinct-cli.py`, and `config.json` files are absent from the bundle, so the commands are not actually runnable as shipped — a key gap that keeps it below the 'mostly executable' anchor. | 3 / 5 |
Workflow Clarity | Quick Start lays out a clear numbered sequence (enable hooks → init directories → use commands) and promotion includes a `--dry-run` preview checkpoint plus explicit criteria (2+ projects, confidence ≥ 0.8). But there are no verification steps for the batch observation pipeline — nothing to confirm hooks are firing, observations are accumulating, or the observer is running — so checkpoints are missing for a batch auto-learning workflow. | 3 / 5 |
Progressive Disclosure | The skill is a monolithic ~350-line SKILL.md with no references/, scripts/, or assets/ directories, yet the body repeatedly points to files (`hooks/observe.sh`, `hooks/hooks.json`, `instinct-cli.py`, `config.json`) that do not exist in the bundle. Content that clearly belongs in separate files — version-history tables, the scope-decision guide, the file-structure reference — is inlined, and the broken paths make navigation unreliable. | 2 / 5 |
Total | 10 / 20 Passed |