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 is an unusually well-organized thin index — lean, sequenced, gated, with a real checklist — and the worked example reference is genuinely useful. Its one serious defect is that the entire `rules/` layer it delegates to is absent from the bundle, so the skill as shipped is a table of contents for files that don't exist, capping actionability and progressive disclosure.
Suggestions
Ship the six `rules/*.md` files the workflow depends on (research-grounding, readiness-gate, consultative-completeness, question-design, brief-artifact, aw-integration) — every phase gate currently points at a dead path.
Merge the "Workflow" and "Required reading by phase" tables: they enumerate the same phase→file mapping twice; a single table with columns Phase / Rule file / Gate / Load-when removes the duplication.
Inline a minimal fallback for the readiness gate (the blocking vs advisory classification rule) in SKILL.md so the skill remains executable even when rule files are absent or stripped.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence — no concept tutorials, just operational tables, principles, anti-patterns, and a definition of done. However, the "Workflow" table and the "Required reading by phase" table list the same six rule files twice, and the "Relation to aw and ideate" section repeats scope already covered in the description, so a few tokens could be trimmed. Not 5: the duplicated phase→file mapping is avoidable redundancy; not 3: nothing here explains what Claude already knows. | 4 / 5 |
Actionability | As written, the body gives concrete orchestration — named phases with gates, "≤ 2 batched rounds" via "AskUserQuestion", the verdict enum, `--non-interactive` flag semantics, and the output path ".agent/{branch}/brief.md". But all six referenced procedure files (`rules/research-grounding.md`, `rules/readiness-gate.md`, `rules/question-design.md`, `rules/brief-artifact.md`, `rules/consultative-completeness.md`, `rules/aw-integration.md`) do not exist in the bundle, so the actual instructions for executing each phase are missing — a reader cannot perform Phase 0–4 from what is present. Not 4: the missing rule files leave key execution details absent, not just minor gaps. | 3 / 5 |
Workflow Clarity | Five phases are clearly sequenced in a table with an explicit gate per phase ("Do not advance until the gate passes"), an adaptive-pass branch ("it asks nothing ... and returns ready"), and a Definition of Done checklist — real checkpoints, and the operations are non-destructive so no validation cap applies. Not 5: the gate detail itself is delegated to rule files that are absent from the bundle, so the validation steps are named but not executable from what ships here. | 4 / 5 |
Progressive Disclosure | The structure as designed is a model thin index — "This SKILL.md is a thin index. Phase procedures live in rules/*.md and load on demand", with phase→file mapping tables and a worked example in `references/interview-example.md` (which exists and is one level deep). But scoring against the actual bundle: the `rules/` directory does not exist, so six of the seven referenced paths are dead links. The disclosure architecture cannot deliver its content, which is a structural gap, not a minor organization issue. Not 4: most referenced content is missing rather than mostly clear; not 2: what is present is well-signaled and appropriately split. | 3 / 5 |
Total | 14 / 20 Passed |