Content
70%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 core workflow (Steps 0-10) is exceptional — fully executable commands, explicit gates, timeouts, and revert paths that set the standard for a production-merge skill. However, roughly a third of the file is inlined framework boilerplate (voice coaching, telemetry/onboarding flows, duplicated detection blocks) that inflates token cost and belongs in shared or separate reference files. Extracting that boilerplate would make an already excellent skill lean and efficient.
Suggestions
Move the generic gstack framework sections (Voice, AskUserQuestion Format, Contributor Mode, Telemetry preamble/footer, Plan Status Footer) into a shared framework reference file (e.g., $GSTACK_ROOT/ETHOS.md or a BASE.md) and keep only skill-specific behavior in SKILL.md — this would cut roughly 400 lines and lift conciseness.
Extract the deploy-config detection bash block (currently duplicated verbatim in Step 1.5a and Step 5) into a single script or one referenced section, and invoke it by reference the second time.
Move the one-time onboarding flows (Boil-the-Lake intro, telemetry opt-in, proactive prompt) out of the critical path of the deploy instructions into a clearly separated 'First-run onboarding' reference so the merge/deploy/verify workflow starts near the top of the file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 1362-line body is noticeably verbose: ~340 lines of generic gstack preamble (telemetry opt-in flows, proactive-mode prompts, 'Boil the Lake' intro), a ~45-line Voice/personality section, and the deploy-config detection bash block duplicated verbatim in Step 1.5a and Step 5. Not a 1 because the core Steps 0-10 are task-specific and free of explaining concepts Claude already knows; the padding is boilerplate and repetition rather than tutorial-style filler. | 2 / 5 |
Actionability | Every step ships copy-paste-ready commands: `gh pr checks --watch --fail-fast`, `gh pr merge --auto --delete-branch`, `fly status --app {app}`, `$B console --errors`, with exact polling intervals, timeouts, and exit conditions. The common cases (GitHub PR + Actions deploy, platform CLIs, auto-deploy platforms) are each covered with executable commands, matching the fully-executable anchor. | 5 / 5 |
Workflow Clarity | Steps 0-10 are clearly sequenced with explicit validation checkpoints: first-run dry-run gate, pre-merge readiness gate (3.5) with review-staleness rules and a rendered report, CI watch with 15-min timeout, merge-queue polling with 30-min timeout, and canary health assessment. Failure paths have feedback loops (STOP-and-fix, revert escape hatch offered at every failure point) — exemplary for an irreversible production-merge workflow. | 5 / 5 |
Progressive Disclosure | Internal structure is good (numbered steps, clear headers, ASCII report templates) and external gstack references like `$GSTACK_ROOT/review/checklist.md` and `$GSTACK_ROOT/gstack-upgrade/SKILL.md` are clearly signaled, but no bundle files exist and everything lives in one monolithic file. The Voice guide, AskUserQuestion format, telemetry protocol, contributor mode, and Plan Status Footer are framework boilerplate inlined here that clearly belongs in shared reference files. Not a 2 because the structure is far from minimal and navigation within the file is easy. | 3 / 5 |
Total | 15 / 20 Passed |