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.
Excellent operational content: fully executable commands, a hard safety invariant (top of stack byte-identical to original) backed by explicit verification at every stage, and validation-first handling of destructive operations. The only structural improvement is moving the large subagent prompt block into a references/ file to slim the main body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and rationale-driven — it explains only non-obvious specifics (why --auto, why not rebase, why no worktree) and assumes Claude's competence throughout. A few asides (e.g. "That failure is the good outcome: it stops a half-split commit") could be trimmed, matching the anchor-4 example of efficient-with-minor-over-explanation. | 4 / 5 |
Actionability | Every step ships copy-paste-ready commands with explicit flags (gh stack submit --auto --open, strip_hunks.py classification.json --dry-run, git checkout <original-sha> -- <files>), plus a fully-specified subagent prompt template — the description and script contract match the actual bundle. | 5 / 5 |
Workflow Clarity | Six clearly sequenced steps with explicit validation checkpoints for destructive/batch operations: --dry-run validation first, the "MUST be empty" git diff check, snippet spot-checks, and specs run on both branches, with feedback loops ("fix the classification, don't patch the child") for error recovery. | 5 / 5 |
Progressive Disclosure | The strip script is appropriately externalized to scripts/strip_hunks.py (a real file, correctly referenced) and the body is well-sectioned and one level deep. Not 5: the ~170-line body is monolithic — the subagent classification prompt template and gh-stack usage notes are natural candidates for a references/ file. | 4 / 5 |
Total | 18 / 20 Passed |