Content
81%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.
This is a well-engineered orchestration skill: fully concrete delegation prompts, commands, and option strings, with explicit approval gates and sign-off feedback loops throughout. Its main weakness is token efficiency — repeated write-immediately directives, long justification blockquotes, and a closing section that duplicates Phase 6 — plus all content living inline with no bundle split despite referencing a docs tree it could lean on further.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, non-obvious pipeline policy rather than filler, but it could be tightened: "Write the approved section to file immediately" repeats in nearly every section, the spawn-policy blockquotes run long ("This is not reduced specialist involvement...", "Separate calls would ask the agent to reason about colour..."), and the closing "Recommended Next Steps" section largely duplicates Phase 6's option pool. It is above anchor 2 (no padding that explains things Claude already knows) but below the 'minor instances' bar of anchor 4. | 3 / 5 |
Actionability | Guidance is copy-paste ready: exact Grep patterns with flags for retrofit detection, verbatim agent delegation prompts, exact AskUserQuestion option strings, exact file paths and verdict-header strings, and explicit handling of edge cases (retrofit vs. fresh, one-page brief fallback, missing project.yaml keys, agent conflicts, solo/lean sign-off skips). Common cases are specifically covered, matching anchor 5. | 5 / 5 |
Workflow Clarity | Phases 0-6 are clearly sequenced with explicit validation at every step: per-section user approval gates before any write, the AD-ART-BIBLE sign-off with feedback loops (REJECT/Revise -> specialist re-draft -> re-approve -> write -> record REVISED; NOT ASSESSED -> name missing input, re-run), a Phase 0 fail-fast check, and a Phase 6 state-check checklist before proposing next steps. This matches the anchor-5 pattern of sequence + explicit validation + error-recovery loops. | 5 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent), and the body is well sectioned with clearly signaled one-level-deep pointers to external docs (.claude/docs/director-gates.md, workflow-modes.md, templates/art-bible.md). It falls short of anchor 5 because ~400 lines of policy — tier tables, the section status-table spec, delegation rationale — are all inline where some could be split into reference files; it is well above anchor 3 since every reference is clearly signaled and structure is strong. | 4 / 5 |
Total | 17 / 20 Passed |