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.
The body is an exceptionally actionable, well-validated workflow: exact commands, explicit stop conditions, decision tables, and feedback loops for every risky operation. Its weaknesses are structural rather than substantive — the Traps table duplicates earlier steps, and the long single file could offload the lockfile procedure to a reference file.
Suggestions
Trim the Traps table to traps not already stated as step rules (e.g. drop the dirty-worktree and main-failure rows, which restate Steps 1 and 4) to cut duplicated tokens.
Move the Step 3 bun.lock regeneration procedure — the change/commands table and evidence-script walkthrough — into a references/ file, keeping a short pointer and the stop conditions in SKILL.md.
Condense the worktree-lifecycle rationale in Step 1 (prunable/no-line/one-line cases) into a compact table matching the style of Steps 3 and 4.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational, assumes competence, and explains only non-obvious repo-specific rationale ("a bare `bun install` drops every entry for a platform this machine is not"), but the Traps table restates roughly six rules already given in Steps 0–4 (e.g. the dirty-worktree and main-failure rules), and a few passages could be tightened. That duplication keeps it at the efficient-with-minor-trims anchor rather than the every-token-earns-its-place anchor. | 4 / 5 |
Actionability | Guidance is copy-paste ready throughout: exact `gh api … --jq` queries, the `bun add '<pkg>@<version>' --cwd <dir>` table with precise flags, regexes for branch-name and package validation, and the evidence-script invocation pinned to main's copy. Decision tables cover the common cases (fork head, prunable worktree, flake vs branch failure), matching the fully-executable anchor. | 5 / 5 |
Workflow Clarity | Steps 0–7 are explicitly sequenced with pervasive validation checkpoints ("must print nothing", "Stop and report", "Anything else → stop and ask the owner"), feedback loops ("A red check goes back to Step 4", "the lockfile is wrong: start this step again"), and a closing stop-and-hand-back checklist. For a risky skill (force-push, rebase, running unreviewed code) the validation coverage satisfies the top anchor rather than the minor-gaps one. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the single body is well-sectioned with one-level-deep, clearly signaled pointers (prep-pr, write-pr, wiki/conventions/contributing.md, scripts/wait-pr.sh, scripts/lockfile-evidence.sh). However, ~230 lines of inline material — notably the full bun.lock regeneration procedure and red-check triage tables — could live in a reference file, which is the minor-organization-gaps anchor rather than the appropriately-split one. | 4 / 5 |
Total | 18 / 20 Passed |