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 exceptionally actionable and the multi-step workflow is sequenced with rare rigor — mechanical sentinels, a hook-enforced plan gate, cache/fingerprint idempotency, and adversarial verification steps. Its weaknesses are length and structure: heavy repetition of the LAW/sentinel mechanism, dated incident narratives, and a monolithic 500-line layout where reference files would keep the overview lean.
Suggestions
Move the three dated 'Named failure mode' narratives and the full LAW exposition into a references/output-contract.md file, keeping one-line summaries plus the sentinel spec in SKILL.md; the sentinel rule is currently restated in at least five places.
Extract the dossier card template, the forged-SKILL.md template, and the authorship/privacy guidance into reference files so the main body is a lean overview of Steps 0–6 with command pointers.
Trim process narration guidance (e.g., the Voice paragraph and repeated 'do not count as a check' reminders) to single imperative lines — the mechanical sentinel already guarantees compliance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The command references and flag tables are dense and useful, but the ~500-line body carries noticeable padding: the LAW 1 sentinel/hook mechanism is restated at least five times (failure mode #3, LAW 1, Step 3, plan-render section, output contract), and the dated incident narratives ("Named failure mode #1 (2026-06-09, BearClaude run)" and #2/#3) retell sessions at length where one line each would do. This fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than anchor 4, where only minor instances would need trimming. | 3 / 5 |
Actionability | Nearly everything is executable: exact engine invocations (`node "<skill-dir>/scripts/mine-skills.mjs" index --dir <history-dir>`, `forged check --emit json`, `theme expand --key <id> --max 40`), a user-input→flag mapping table, the manifest JSON shape, the canonical-then-symlink install loop, and a complete forged-SKILL.md template with verification and failure-mode sections. All four referenced script paths exist in the bundle. Not 4: commands are copy-paste ready and cover the common cases end to end. | 5 / 5 |
Workflow Clarity | Steps 0–6 are clearly sequenced with explicit validation checkpoints and feedback loops: the Step 2b truth check (re-open evidence, weigh outcomes, refute the cheap explanation), fingerprint-matched dossier caching ("If the cached fingerprint matches ... reuse it and skip mining"), registry-driven up-to-date/update/suppress decisions, confirmation before the destructive `reset`, and the mechanically checkable sentinel plus PreToolUse hook that rejects non-conforming plans with instruction to re-render. Not 4: error-recovery behavior is explicit (hook denies → render manifest and re-present; cancellation recoverable via `plan last`). | 5 / 5 |
Progressive Disclosure | The bundle is well-used for executable content — mine-skills.mjs, deep-mine.workflow.js, theme-sweep.workflow.js, and hooks/validate-plan.mjs are all real, one-level-deep, and clearly invoked — but everything else is inlined in a single ~500-line body: the three incident narratives, the LAW treatise, the authorship/privacy guidance, the dossier card format, and the forged-skill template are all prose blocks that belong in reference files. This sits at 'some structure but could be better organized; content that should be separate is inline' rather than anchor 4, where organization gaps would be minor. | 3 / 5 |
Total | 16 / 20 Passed |