Content
70%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 well-sequenced orchestration contract with excellent validation gates and checklists, but its progressive-disclosure design collapses in practice because the docs/ and prompts/ files it delegates all commands and detailed rules to are missing from the bundle, leaving the body itself without executable commands. Shipping the referenced files or inlining the essential script invocations would resolve both weaknesses.
Suggestions
Include the referenced 'docs/' and 'prompts/slide-worker.md' files in the skill bundle — the body delegates all commands and detailed rules to them, and their absence makes every 'read docs/...' instruction a dead reference.
Inline at least the minimal executable invocations for the bundled scripts (e.g. how to call record_slide_result.py, prepare_slide_prompts.py, assemble_ppt.py with their arguments) so the workflow is executable without the docs/ layer.
Consolidate the repeated subagent-mandate and backend-fixity statements (currently restated in Hard Constraints, workflow steps, Subagent Dispatch, and Acceptance Criteria) into one authoritative section to trim tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is imperative and dense with no explanation of concepts Claude already knows — no padding about what PPT or image generation is. It sits at 4 ('minor instances of over-explanation that could be trimmed') rather than 5 because the same constraints are restated across sections: the subagent mandate appears in Hard Constraints, steps 8-9, Subagent Dispatch, and Acceptance Criteria; backend fixity and 'don't mark complete from chat alone' are each repeated three-plus times. | 4 / 5 |
Actionability | Concrete artifacts are named throughout ('scripts/prepare_slide_prompts.py', 'record_slide_result.py', 'origin_image/slide_XX.png', state values 'pending/dispatched/recorded/accepted'), but the body contains no executable command or invocation syntax for any script — all commands are delegated to 'docs/slide-generation-and-subagents.md' and other docs/ files that are absent from the bundle. That is 'some concrete guidance but incomplete; missing key details', not the mostly-executable level of score 4. | 3 / 5 |
Workflow Clarity | The 13-step workflow is clearly sequenced with explicit validation gates ('If any slide is pending, dispatched, or blocked, stop'), a QA feedback loop (inspect every slide, regenerate severe failures, validate edited output before replacing), a user-visible progress checklist tied to real file evidence, and blocker reporting with slide id and evidence. This matches the score-5 anchor: explicit validation steps, error-recovery loops, and checklists for a complex batch process. | 5 / 5 |
Progressive Disclosure | The structure is well designed — a Reference Map with per-file descriptions and read-timing guidance ('read only when config is needed'), one level deep — but scoring against the actual bundle, the 10 referenced 'docs/*.md' files and 'prompts/slide-worker.md' do not exist, so the primary detailed content the body defers to is unreachable dead references. Only 'references/*.md' and 'scripts/' are actually present. Good signaling with broken navigation lands at 3, above 'references buried' (2) but below the reliable one-level-deep organization of 4-5. | 3 / 5 |
Total | 15 / 20 Passed |