Content
20%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is a heavily process-oriented document that prioritizes governance and safety rails over actionable guidance. It reads more like an organizational policy document than a skill that teaches Claude how to perform a specific task. The extreme abstraction, domain-specific jargon without definitions, and lack of concrete examples (templates, code, sample outputs) make it very difficult to follow or execute.
Suggestions
Add a concrete example of a complete reframe program output (even abbreviated) showing the expected schema_version, phases, rollback conditions, and eval proof structure so Claude knows exactly what to produce.
Replace abstract procedure steps like 'Resolve the subagent stage map from routing-map.json' with concrete examples showing what the resolution looks like and what decisions result from it.
Consolidate the repeated 'Do Not Create' criteria into a single decision table or checklist rather than scattering the concept across 6+ sections.
Remove or drastically shorten sections that restate the same constraints (Constraints, Execution Boundaries, Failure Mode, Gotchas, and Anti-Patterns all overlap significantly) and move detailed governance rules to a referenced contract file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is extremely verbose and filled with domain-specific jargon that reads like process theater. Many sections repeat similar constraints (e.g., 'Do Not Create' is mentioned 6+ times), and concepts like 'architecture-evolution compression,' 'stage arc boundary,' and 'source-prompt family status' are never concretely defined. The content could be reduced by 50%+ without losing actionable information. | 1 / 3 |
Actionability | Despite its length, the skill contains no executable code, no concrete command examples (except two validation commands buried at the end), no template showing what a reframe program actually looks like, and no specific data structures. The procedure steps are abstract directives ('Resolve the subagent stage map,' 'Apply the BLUF review contract') that assume extensive undocumented context without providing concrete guidance. | 1 / 3 |
Workflow Clarity | The 12-step procedure provides a numbered sequence with some validation checkpoints (steps 7, 12) and failure modes. However, the steps are abstract and heavily cross-referenced to external contracts without explaining what passing or failing looks like. The validation section does include specific commands and a 'fail fast' directive, which partially redeems the workflow clarity. | 2 / 3 |
Progressive Disclosure | The References section provides well-organized one-level-deep links to external contracts and references, which is good structure. However, since no bundle files are provided, we cannot verify these references exist. The main body itself is monolithic and could benefit from splitting detailed procedure steps, constraints, and anti-patterns into separate reference files rather than inlining everything. | 2 / 3 |
Total | 6 / 12 Passed |