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.
The body is a tightly written, highly actionable repair loop: concrete gh commands, a failure triage table, explicit state classification, and a genuine verify-on-remote feedback loop. Its weaknesses are modest — some redundant GitHub-state enumeration, placeholder variables in the GraphQL example, and inline reference-grade material that a ~200-line skill could offload to a references/ file.
Suggestions
Trim or compress the enumerated check-run/status state lists to the non-obvious ones (e.g., STALE, STARTUP_FAILURE) since Claude already knows the common states.
Define $PR_NUMBER, OWNER, and REPO before the snippets that use them so every command block is copy-paste ready.
Move the GraphQL statusCheckRollup query into a short references/ file to keep the main workflow lean.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Largely lean and command-driven with no padded conceptual explanation, e.g., direct `gh pr view --json ...` and `gh run view <RUN_ID> --log-failed` snippets. However, the full enumeration of GitHub states ("QUEUED, PENDING, WAITING, REQUESTED, IN_PROGRESS") repeats knowledge Claude already has, so it is not anchor-5 lean. | 4 / 5 |
Actionability | Mostly executable guidance: concrete gh/jq commands, a failure-classification table mapping type to action, and explicit push/poll steps. Minor gaps keep it below 5 — the GraphQL snippet uses literal OWNER/REPO placeholders and `$PR_NUMBER` is referenced in step 2 before its derivation is shown. | 4 / 5 |
Workflow Clarity | A clear 8-step sequence with explicit checkpoints and feedback loops: poll-until-terminal criteria, terminal/pending/failure state definitions, restart points ("restart from Step 3"), explicit exit conditions, and a remote verification gate ("The loop is only complete when the remote PR checks for the new head SHA are green"). Not a destructive/batch operation, so no cap applies. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the single-file body is well-sectioned (Scope, Inputs, Workflow, Output, Notes) with no nested references. At ~200 lines, content such as the inline GraphQL payload and the check-state tables is a candidate for a one-level-deep references file, which is the minor organization gap that keeps it below anchor 5. | 4 / 5 |
Total | 17 / 20 Passed |