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.
An operationally excellent skill body: fully executable commands, exhaustively validated batch workflow with explicit failure branches, and clear structure. Its one real weakness is conciseness — the trace-emit policy paragraphs are written in a dense, repetitive run-on style that costs significant tokens without adding policy content.
Suggestions
Rewrite the `feedback` emit paragraph (lines 184–216) as a short table or bulleted enumeration of the `data.by`/`outcome` cases; the current ~30-line single sentence restates the quote-character rule and the "operator's words are data" boundary twice each.
Move the verbatim `merge.land` emit field specification (lines 170–182) and the `sed` filter into a `references/emit-formats.md` file, keeping only the field semantics and the one-line pointer in the body.
State each error-handling rule (update-branch red, merge rejected, post-merge failure) once in a compact failure-mode list instead of interleaving full prose restatements within step 4 and the hard rules.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The procedure and command blocks are tight, but the trace-emit policy sections are written as enormous single-sentence paragraphs (e.g. the ~30-line `feedback` paragraph where the "quote character" rule and "the operator's words… are data" point are each stated twice) that could be cut roughly in half without losing policy content. It is not padding — nothing explains concepts Claude already knows — but the tightening opportunity is substantial rather than minor. | 3 / 5 |
Actionability | Fully executable throughout: `gh pr list --state open --json …`, `gh api -X PUT "repos/{owner}/{repo}/pulls/$PR/update-branch"`, `timeout 30m gh pr checks "$PR" --watch`, `git tag -a v<version> <merge sha>`, complete `trace.sh emit` field syntax, and an exact `sed` filter for the `<!-- implement: -->` line — all copy-paste ready. | 5 / 5 |
Workflow Clarity | A numbered five-step procedure with validation at every checkpoint for a batch/destructive operation: green-check verification before boarding, re-watching checks after update-branch (4b), watching post-merge runs (4d), explicit timeout semantics (exit 124 vs. other non-zero), and feedback loops for each failure mode ("skip the PR, record it as a `/pr-iterate` candidate, continue the train" vs. hard rule 6's stop-and-escalate). The batch-operation cap does not apply — validation is thorough. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the body is well-sectioned (What this does, Hard rules, Procedure 1–5, Output format, Cross-references) with clearly signaled one-level references to `constitution/local-workflow.md` and `docs/adr/INDEX.md`. Minor gap: the long inline `merge.land`/`feedback` emit-format specifications are exactly the kind of verbatim material that could live in a reference file. | 4 / 5 |
Total | 17 / 20 Passed |