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.
An unusually rigorous deployment skill: the orchestration body stays lean by delegating host detail to per-host reference files, provides fully executable commands, and wraps destructive batch operations in extensive validation gates and recovery paths. The main issues are a dangling UPGRADING.md reference and a few run-on paragraphs that would benefit from restructuring.
Suggestions
Remove the UPGRADING.md row from the reference table or ship the file — it is referenced but absent from the bundle, so following it fails.
Break the Phase 1 self-check paragraph and the .story-deployed target_cli rules into short bullet lists; each currently packs 5+ conditional rules into one run-on sentence.
The host list (claude-code, opencode, codex, antigravity, zcode, openclaw, reasonix, generic) is enumerated three times (step 5 markers, step 6 answers, sentinel target_cli); state it once and reference it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is densely packed with project-specific operational detail and assumes Claude's competence (no explanations of basic concepts), but several oversized paragraphs — the Phase 1 self-check paragraph and the sentinel target_cli rules — cram many rules into single run-on sentences and repeat the host enumeration, and could be split or tightened. This is efficient with minor trimming opportunities, not score-5 lean. | 4 / 5 |
Actionability | Guidance is fully executable: exact file and directory names to verify, a copy-paste sentinel field block with `date -u +"%Y-%m-%dT%H:%M:%SZ"`, the complete `merge-project-instructions.py` command with every flag and placeholder defined, the interpreter probe order (python3 → python → py), and numbered manual-merge fallback rules. | 5 / 5 |
Workflow Clarity | The three-phase sequence is explicit with strong validation checkpoints: pre-deploy bundle self-check with a hard stop, version gates that stop to avoid downgrade, copy-path-safety realpath/samefile checks before every recursive copy, per-host validation sections, Phase 3 sentinel verification, and an idempotent re-run-from-start recovery path for interruptions. Despite destructive/batch operations (pruning, overwriting managed files), validation and feedback loops are thoroughly specified. | 5 / 5 |
Progressive Disclosure | The SKILL.md is a well-structured orchestration overview that delegates all host-specific detail to eight one-level-deep references/deploy-*.md files (all present) with an explicit "read only the target host's file" instruction, and every referenced script and templates/hooks/lib/sentinel.sh exists. However, UPGRADING.md is cited in the reference table but is missing from the bundle, a dangling reference that prevents a top score. | 4 / 5 |
Total | 18 / 20 Passed |