Content
96%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 exemplary skill body: a 13-step workflow where each step produces a concrete, example-backed artifact, with explicit validation gates, escalation paths, an approval feedback loop, and an anti-patterns section that converts vague directives into rubric assertions. Token efficiency is excellent — dense tables and telegraphic directives with zero teaching of known concepts. The single structural gap is that all detail lives inline in one file with no separate reference files for reusable templates (recovery matrix, eval policy, slot architecture).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and telegraphic throughout — tables, ASCII diagrams, code blocks, and compressed directives like "SSoT: personality_profiles.system_instruction (per-tenant)" and "Métricas en bullets. NUNCA dumps largos." It assumes Claude's competence (never explains what LangGraph or prompt caching are) and every section carries project-specific instruction rather than padding. This matches the lean-and-efficient anchor; it is not score 4 because there is no over-explanation to trim, only minor deliberate reinforcement between the cache-forbidden list and anti-patterns. | 5 / 5 |
Actionability | Every step ships a concrete, filled-in artifact: an example turn-by-turn transcript, an ASCII state machine with timeouts, a tools table with real signatures, a slot layout with TTLs and invalidation triggers, a recovery matrix with detection thresholds ("sentiment grader < 0.3"), numeric budgets ("TTFT p95: <2s"), and copy-paste-ready iteration/handoff/checkpoint message templates. For an instruction-only design skill this is the fully executable equivalent, matching the top anchor; it is not score 4 because the specific examples cover the common cases with no key details missing. | 5 / 5 |
Workflow Clarity | Thirteen clearly sequenced steps with explicit validation checkpoints and feedback loops: Step 1 gates on a clear contract ("Si contract no claro → escala /po"), Step 11 escalates discovered edge cases via delta-spec.md → /po, and Step 12 iterates "Loop hasta aprobación" before handoff. This matches the anchor for clear sequence with explicit validation steps and feedback loops; it is not score 4 because checkpoints and escalation paths are explicit rather than minor-gapped, and the destructive/batch cap does not apply to this design workflow. | 5 / 5 |
Progressive Disclosure | The file is well-sectioned with clear headers (Diferencia, Inputs, Workflow steps 1-13, Anti-patterns, Output format) and easy in-file navigation, and no references are buried. However, no bundle/reference files exist: all detail (recovery matrix, eval policy format, slot architecture templates, example transcripts) is inline in a single ~245-line file with no one-level-deep references splitting content out. This matches the good-structure/minor-organization-gaps anchor; it is not score 5 because the top anchor expects well-signaled separate reference files, and not score 3 because the inline placement is reasonable and clearly organized rather than disorganized. | 4 / 5 |
Total | 19 / 20 Passed |