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.
A high-quality, executable runbook for a fragile destructive operation: clear sequencing, explicit validation and feedback loops, and a strong guardrail checklist. Its only real weaknesses are minor conciseness padding and the absence of any progressive-disclosure split, which is largely justified here by the self-contained workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense procedural guidance that mostly earns its tokens, but includes minor over-explanation Claude does not need (the worked conflict-marker block at lines 105-112) and repeats the STATE RB-MERGE reminder nearly verbatim at lines 34 and 158. | 4 / 5 |
Actionability | Commands are fully executable and copy-paste ready given the dispatch payload (e.g. "cd <WORKTREE_PATH> && git revert --no-edit -m 1 <sha>", "git show --stat <sha>"), and the Outcome A/B1/B2/C branching covers the common cases concretely. | 5 / 5 |
Workflow Clarity | A clear Step 0-4 sequence with explicit validation checkpoints (re-check "git diff --name-only HEAD^ HEAD" after every revert), feedback loops (revert your last edit on over-removal; re-apply Outcome B logic after a revert --continue), and a NEVER checklist for a destructive/batch operation. | 5 / 5 |
Progressive Disclosure | The body is well-organized with clear ##/### headers (Spawn prompt, Output contract, Workflow, NEVER) and is appropriately self-contained with no external references needed, though it is a single monolithic file rather than an overview pointing one level deep into bundled references. | 4 / 5 |
Total | 18 / 20 Passed |