Content
96%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 exceptionally dense, operational skill: every workflow is executable with explicit validation, error-recovery via exit codes, and confirmation gates on batch/destructive actions. The only structural nit is that reference tables (exit codes, TUI workarounds) live inline rather than in bundle files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body stays lean and assumes Claude's competence: every section teaches gh-stack-specific behavior Claude cannot know (e.g. "`sync` exits 0 with 'Sync aborted' when the local and GitHub stacks diverge", exit codes 2/3/7/8/10, the rerere pre-setup) rather than re-explelling general git or PR concepts. Nothing reads as padding; the one general-knowledge touch ("GitHub refuses self-approval") is a one-clause justification for the script's behavior, not an explanation. | 5 / 5 |
Actionability | Every instruction is a copy-paste-ready command with its arguments and exit semantics: `bash <skill-dir>/status.sh [pr-number | pr-url | branch]`, `gh stack submit --auto --open`, `gh stack rebase --continue # repeat until it exits 0`, `gh pr edit <pr> --title <title> --body-file <file>`, `git config rerere.enabled true`. The common cases (status, sync, watch, conflict, approve, build) each have a complete runnable path, including flag variants (`--interval`, `--timeout`, `--grace`, `--prune`, `--dry-run`). | 5 / 5 |
Workflow Clarity | The update flow is a numbered sequence with explicit validation and feedback loops: rerun `status.sh` and show it as "the proof the sync landed", `watch.sh` polls until checks settle, and the conflict loop is resolve → `git add` → `rebase --continue` until exit 0 → `push` → rerun the report. Destructive/batch operations are gated: "Confirm before these" requires echoing scope and merge method back to the user, and approve distinguishes the go-ahead from status-only requests via `--dry-run`, so the batch/destructive cap does not apply. | 5 / 5 |
Progressive Disclosure | Structure is good: three one-level-deep scripts (`status.sh`, `watch.sh`, `approve.sh`) are clearly signaled with usage lines and comments, and flag lists are deferred to `gh stack <command> --help` instead of being inlined. Minor gap: the exit-codes table and the TUI-trap details are reference-style material inlined in SKILL.md, and there is no `references/` or `scripts/` bundle listing to aid discovery of the scripts beyond their in-text mentions. | 4 / 5 |
Total | 19 / 20 Passed |