Content
88%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.
An unusually strong workflow skill: fully executable commands, an explicitly ordered dependency chain with real validation gates (stress-plan, the gate probe, NOT RUN tracking), and almost no wasted tokens on things Claude already knows. The residual costs are density — long stretches of rationale-prose in the operative steps — and a body that keeps reference-shaped material inline rather than splitting it out.
Suggestions
Conciseness: trim the justificatory prose in Steps 0 and 2 to bare rules — e.g. compress "a wrong config value produces a tidy diff, a green build and no behaviour change, which no gate can fail on" to "a wrong config value fails no gate", and cut the session-metrics rationale for the complexity rating to one clause.
Progressive disclosure: move the issue body template and the Step 2 plan-content checklist (wiki-staleness, config-value citations, observability requirements) into a references file, leaving SKILL.md as the sequenced overview with clearly signaled one-level-deep pointers.
Progressive disclosure: add a short list at the top of the external paths the skill relies on (references/skills-routing.md, wiki/conventions/wiki-search.md, wiki/shared/observability/overview.md, scripts/changeset-required.sh) so navigation does not depend on reading Steps 2–4.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Almost every token carries repo-specific information Claude cannot know — bare-repo paths, the PERSONAL/REMOTE probe, label taxonomy, "@cire/* packages are version-less" — with no explanation of concepts Claude already knows. Not 5: several rationale-heavy passages ("a wrong config value produces a tidy diff, a green build and no behaviour change, which no gate can fail on"; the rate-complexity "contaminated and worthless" justification) run longer than the instruction needs. | 4 / 5 |
Actionability | Guidance is copy-paste ready: the `gh issue create` command includes the full four-field body heredoc, the worktree and `git checkout -B` commands are complete, and the upfront probe script is executable verbatim. Not 4: the few soft spots (project item-edit, the Plan-subagent prompt) still name their exact command or required opening line and a fallback. | 5 / 5 |
Workflow Clarity | The sequence is explicit and dependency-ordered ("Four things, in this order — each depends on the one before"; Steps 0–4 with the Finish checklist), and validation is built in: the pre-flight gate probe, the stress-plan attack with "every finding... fixed in the plan or rejected in writing", and the NEW-FEAT.md convention of recording unrun gates as "NOT RUN". Not 4: checkpoints and feedback loops are explicit, not implicit. | 5 / 5 |
Progressive Disclosure | The one bundle file, references/skills-routing.md, exists, is exactly one level deep, and is clearly signaled in Step 4 ("The routing table is `references/skills-routing.md`"); detail on observability and wiki search is deferred to wiki pages rather than inlined. Not 5: the body itself is a dense ~140-line monolith with no navigation/signposting of its external references, and inlines material (the issue body template, the plan-content checklist) that could live in reference files. | 4 / 5 |
Total | 18 / 20 Passed |