Content
63%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 delivers strong, concrete operational guidance — complete output templates, an explicit single-clarifier discipline, a hard no-fabrication rule with fallback phrasing, and an approval gate — but it is undermined by two systemic issues: roughly a third of its structure duplicates the frontmatter description or repeats itself across sections, and all six bundle files it depends on for detail (detection tactics, voice anti-patterns, complexity worked examples, and three helper scripts) are missing from the skill directory.
Suggestions
Delete the Invocation Triggers section (or reduce it to the implicit-signal bullets): the frontmatter description already lists all nine trigger phrases verbatim and is always loaded with the skill, so the body copy is pure token duplication.
Ship the referenced bundle files (references/workspace_detection.md, references/voice_preservation.md, references/complexity_matching.md, and the three scripts) or strip the references — as written, Section 3's deterministic inventory step, the compressed-format trigger, and three detail lookups all dead-end on missing files, which is the core progressive-disclosure failure.
Merge the Error Handling table into the inline rules it restates (workspace inaccessibility, "go" before approval, short dumps), reconcile the compressed-format threshold ("5 or fewer items" vs "3-5 items"), and drop the maintainer footer (source spec / build pattern / gitignored-provenance) from the distributable skill.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The guidance itself is dense and assumes Claude's competence (no basic-concept explanations), but there is real duplication: the Invocation Triggers section re-lists all nine frontmatter trigger phrases verbatim, "fast-to-action / no upfront intake / at most one clarifier" is stated in both the intro and the Grill-Me section, and the Error Handling table restates rules already given in Section 3 and the Approval Gate. The maintainer footer (source spec, build pattern, gitignored-megaprompt provenance) also spends tokens without aiding execution. Mostly efficient but could be tightened — not 2, since no section is mere padding; not 4, since the duplication spans several sections and is more than a minor trim. | 3 / 5 |
Actionability | Concrete, copy-paste-ready output templates for every format ("### {Project name in user's voice}" with Q:/Decide: lines, the compressed "## What I heard" template), a right-pattern/anti-pattern table with output paths ("Output: docs/consensus-options.md"), and exact commands ("python workspace_inventory.py --root . --keywords "k1,k2""). Not 5 because every referenced helper is missing from the bundle: the three scripts and three reference docs do not exist, so "Use scripts/workspace_inventory.py to do this deterministically" and the complexity_estimator trigger are not executable as written. Not 3 because the core deliverable guidance is fully specified inline and the scripts have explicit judgment fallbacks. | 4 / 5 |
Workflow Clarity | The flow is clear and well-gated: trigger → at most one clarifier (with an explicit stop condition, "Max 1 clarifying question per dump") → sections → directive question → Approval Gate ("Wait for the user's explicit pick"). Section 3 has an explicit numbered 4-step workflow with a hard no-fabrication rule and fallback responses. Not 5 because there is no single explicit end-to-end sequence (the ordering of Sections 1-4 is implied), and the compressed-format trigger is inconsistent: "5 or fewer items" in the Compressed Output Pattern vs "Dump is very short (3-5 items)" in Error Handling. Not 3 because checkpoints (approval gate, stop condition, error table) are explicit; this is not a destructive or batch operation, so no validation cap applies. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned and its references are clearly signaled and one level deep ("See references/workspace_detection.md for the per-context detection-tactic catalog", a References section with per-file purpose statements, a Tooling table for the scripts). However, scored against the actual bundle structure, none of the six referenced files (references/workspace_detection.md, voice_preservation.md, complexity_matching.md; scripts/workspace_inventory.py, dump_classifier.py, complexity_estimator.py) exist — the disclosure chain dead-ends at every link. Not 4 because broken references are more than a minor organization gap; not 2 because the body itself is well structured and mostly self-sufficient via judgment fallbacks. | 3 / 5 |
Total | 14 / 20 Passed |