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 an executable two-stage review pipeline with concrete bash, status markers, decision tables, and a stage gate. It is padded with mandatory-contract boilerplate and rationale prose, and leans on external scripts not present in this skill's bundle with no in-skill progressive disclosure to offset the inlined scripts.
Suggestions
Trim the 'Execution Contract (MANDATORY)' / 'PROHIBITED' boilerplate and the multi-LLM rationale prose to reduce tokens without losing the enforced workflow.
Move the stub-detection and PR-posting bash into scripts/ files the body invokes, turning the body into an overview that points one level deep.
Proceduralize the fix-or-override and multi-LLM synthesis feedback loops as explicit numbered steps or checklists to push workflow clarity higher.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient with concrete bash and tables, but includes trimmable boilerplate (the 'Execution Contract (MANDATORY)' and 'PROHIBITED' blocks, host-adapter note) and rationale prose Claude already knows, matching the 'mostly efficient but could be tightened' anchor. | 3 / 5 |
Actionability | Provides copy-paste-ready bash for intent loading, the five stub-detection checks, git diff acquisition, PR detection, and PR commenting, with only minor gaps where the multi-LLM dispatch and quality-review dimensions lean on unverified external tooling or checklists rather than executable code. | 4 / 5 |
Workflow Clarity | A clear two-stage sequence has an explicit Stage 1 gate, status markers, decision tables, and an error-handling table; validation checkpoints are present, with only minor gaps where the multi-LLM synthesis and fix-or-override loops are described but not fully proceduralized. | 4 / 5 |
Progressive Disclosure | The skill has no bundle files and inlines substantial scripts (stub detection, PR posting) that a well-disclosed design would split into scripts/, while its external references (codex-host-adapter.md, orchestrate.sh) are unverified and buried in code rather than clearly signaled one-level-deep in-skill references. | 3 / 5 |
Total | 14 / 20 Passed |