Content
77%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.
The content is a model of token efficiency and the 12-phase workflow is clearly sequenced with real validation checkpoints and feedback loops. It is let down by execution gaps: the pivotal Workflow-engine wiring example is left as commented pseudocode, and the bundle is missing the engine, roster, and persona files the instructions depend on, so a reader cannot actually run the skill end-to-end from what ships. Strong on paper, incomplete in practice.
Suggestions
Ship the referenced files or remove the references: engine/team-workflow.engine.js, roster.json / default-roster.json, and personas/<archetype>.md are named as the skill's core mechanism but are absent from the bundle — either include them or rewrite the phases to work without them.
Make the Workflow snippet executable: replace the commented agent() body and '<abs path>' placeholder with real code that reads subagent_type/tier from roster.json and calls agent(), or explicitly justify why the wiring must be filled in by the reader.
Add a fallback path for when the Workflow tool or engine is unavailable (Phase 2 preflight currently only checks readability), e.g. how to run the four helper scripts standalone to salvage partial value from the skill.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — it never explains what subagents, DAGs, or retry loops are; every phase is compressed to its decision-relevant content (e.g. "Phase 7 — Retry / bounce (W): N=2 retries with feedback; the two-track cutoff ... decides ship-vs-regress"). It earns anchor 5 ('every token earns its place'); the only candidates for trimming are the two v1-deferred caveats, but both carry operational weight (do not design competitive nodes in v1), so this is not the 'minor instances of over-explanation' of anchor 4. | 5 / 5 |
Actionability | The helper-script guidance is copy-paste ready ("echo '{...}' | npx tsx scripts/<name>.ts", plus a fully instantiated roster-postmortem input example), but the centerpiece Workflow snippet is pseudocode: the agent() binding body is a comment ("// Build the subagent prompt from node.prompt + the archetype persona ...") and the import path is an "<abs path>" placeholder. This matches anchor 3 ('pseudocode instead of executable code; missing key details') rather than 4, because the primary execution path cannot be run as written and the persona/roster files it depends on are not in the bundle. | 3 / 5 |
Workflow Clarity | The 12 phases are explicitly sequenced with O/W ownership and validation checkpoints throughout: Phase 5 validates the RunSpec ("spec_invalid → fix and retry"), Phase 7 is a feedback-driven retry loop with a two-track cutoff, Phase 8 enumerates error-recovery options on escalation, and Phase 10 gates shipping on artifact-contract-check plus tests per tests_posture. This matches anchor 5 (clear sequence, explicit validation, feedback loops for error recovery); the batch-operation cap does not apply because batch dispatch is exactly what Phase 10 validates. | 5 / 5 |
Progressive Disclosure | The SKILL.md is a genuine overview with well-signaled one-level-deep pointers (references/escalation-contract.md, references/run-spec-schema.md, references/scoring-protocol.md, and four scripts/*.ts — all real, substantive files), but scored against the actual bundle several core referenced paths do not exist: engine/team-workflow.engine.js (the central mechanism, referenced three times), roster.json, default-roster.json, and personas/<archetype>.md. Structure and signaling are better than anchor 3, but the missing heart-of-the-system files are more than the 'minor organization gaps' of anchor 4 — navigation breaks precisely where the reader needs to go deepest. | 3 / 5 |
Total | 16 / 20 Passed |