Content
62%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 concrete and well-checkpointed, with strong validation gates and templates, but it is held back by substantial internal repetition and a monolithic structure that keeps the framework detail inline instead of splitting it into reference files.
Suggestions
De-duplicate the profile contract: state role/round/section counts once (e.g. in 'Required Output Sections (By Profile)') and have 'Output Contract Guardrail' and 'Controlled Execution Profiles' reference it rather than restating it.
Move the lengthy 'Blind-Spot Gated Overlay Workflow' into a dedicated reference file (e.g. references/blind-spot-gated.md) and keep only the invocation surfaces and a one-paragraph summary in SKILL.md.
Merge 'When NOT to Use This Framework' with 'Applicability Pre-Check (Gate 0)' since they cover the same skip-criteria, to remove the overlap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~440-line body is noticeably verbose through internal repetition: the profile contract (role/round/section counts) is restated in 'Output Contract Guardrail', 'Controlled Execution Profiles', and again fully in 'Required Output Sections (By Profile)', and 'When NOT to Use This Framework' overlaps with 'Applicability Pre-Check'. It does not pad with concepts Claude already knows, but several sections are redundant and could be tightened. | 2 / 5 |
Actionability | Guidance is highly concrete — exact role/round/section counts, numeric scoring gates (≥5/6, ≥2/2), mandatory confidence-tag syntax, a copy-paste Post-Use Self-Check template, and per-failure recovery actions. It stops short of a 5 only because no fully worked roster/dialogue example is shown inline (those live in references). | 4 / 5 |
Workflow Clarity | A clearly sequenced 7-step Workflow is reinforced by explicit validation gates (Applicability Pre-Check, Real-Person Scoring Matrix, Roster Diversity Matrix), 8-item preflight and 4-item postflight checklists, per-round Rolling Uncertainty Tracker, an Anonymous Meta-Review step, and Failure Modes with recovery actions — matching the top anchor for checkpoints and feedback loops. | 5 / 5 |
Progressive Disclosure | Bundle files are well-organized and real (references/ holds eval-rubric, eval-cases, first-use-guide, and en/zh output-templates; scripts/ holds the lint/run validators), and the body signals them with backtick paths. However SKILL.md itself is monolithic — the full four-lens framework, repeated profile contracts, and the long blind-spot-gated overlay workflow are inlined rather than offloaded to reference files, so it reads as more than an overview. | 3 / 5 |
Total | 14 / 20 Passed |