Content
81%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.
This is a highly actionable, well-sequenced orchestration skill: every phase carries exact commands, prompts, and templates, with validation and feedback loops throughout. Its weaknesses are repetition and rationale padding that could be trimmed, and the absence of any bundle files to split its considerable inlined detail into.
Suggestions
Deduplicate repeated instructions — the "read by headings with Grep (pattern ^#) then Read ranges" guidance and the "never end with plain text" rule each appear twice; state each once and reference it.
Move the full review-log entry format and the multi-paragraph re-review skip policy into a reference file (e.g. references/review-log-format.md), keeping only the essentials inline.
Trim rationale passages (the NOT ASSESSED verdict explanation, the warnings about partial re-reviews) to their one-line operative rules to improve token efficiency.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by dense, non-obvious procedural guidance, but it contains genuine redundancy — the Grep/Read-by-heading instruction and the "never end with plain text" rule each appear twice, and the re-review skip policy and NOT ASSESSED verdict carry multi-paragraph rationale that could be tightened. | 3 / 5 |
Actionability | Fully executable throughout: exact Bash invocations (gdd-structure-check.sh, review-receipts.sh check/hash), flag-level grep patterns (-A 6, -B 8, [_ ] matching), ready-made AskUserQuestion prompts with option strings, a copy-paste review-log template, and a complete output format. | 5 / 5 |
Workflow Clarity | Phases 0–5 are clearly sequenced with explicit validation checkpoints (freshness receipt check first, structure-check script, AC testability flags), feedback loops (escalate to full re-review on conflict, post-revision re-review options), a section checklist, and defined handling for the un-reviewable case (NOT ASSESSED), with all file writes gated through user prompts. | 5 / 5 |
Progressive Disclosure | Phases are well-organized and external materials are one level deep with clear read/no-read signals ("the spawned agent reads its own gate file; do not read it in the parent session"), but the skill ships as a single ~470-line monolith and inlines detail (the long review-log format, the skip-policy rationale) that belongs in its own reference files. | 4 / 5 |
Total | 17 / 20 Passed |