Content
92%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a dense, fully-actionable runbook with clear sequencing and validation for a destructive multi-shape operation. Its only real gap is progressive disclosure: four referenced files are signaled but not present in the bundle.
Suggestions
Include the referenced files under references/ (changesets.md, revisions.md, filters-and-queries.md, cub-cli.md) so the signaled one-level-deep navigation actually resolves.
If any referenced file is not essential, remove its entry from the References section to avoid pointing at missing material.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Assumes Claude knows the domain (Unit, ChangeSet, head, promote) and never pads with conceptual explanation; every section (preflight, targets, Shapes A/B/C, verify, stop conditions, tool boundary) is operational and earns its place. | 3 / 3 |
Actionability | Provides fully executable, copy-paste-ready bash for each shape with real flags (--restore, --patch, --tag, --filter, --merge-source) and concrete preflight/verify commands; placeholders like <space> are standard parameterization, not pseudocode. | 3 / 3 |
Workflow Clarity | Multi-step flows are explicitly sequenced (preflight gates 1–5, Shape B steps 1–3, verify chain 1–4) with validation checkpoints (cub unit diff to confirm head, bridgestate Ready check) and a Stop conditions feedback loop for a destructive/batch operation. | 3 / 3 |
Progressive Disclosure | The body cleanly signals one-level-deep references (references/changesets.md, revisions.md, filters-and-queries.md, cub-cli.md) with a dedicated References section describing each, but none of those files exist in the bundle, so navigation is not actually resolvable. | 2 / 3 |
Total | 11 / 12 Passed |