Content
85%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 exceptionally actionable, well-structured instruction-only skill: exact question templates, output format strings, bias recognition cues, and a mandatory closing taxonomy leave nothing ambiguous, and bundle files are real, well-signaled, and one level deep. The one real weakness is redundancy — the trigger list, Anti-Patterns section, and Stop Directive rationale duplicate content already stated elsewhere in the skill or in the description.
Suggestions
Trim the Invocation Triggers list in the body to a short pointer ('triggers listed in the description') or a few examples, instead of repeating all 11 phrases already present in the frontmatter description.
Delete or collapse the Anti-Patterns section — roughly five of its seven items restate rules already given in Tone and Format Rules and Error Handling.
Cut the 'This matters because' bullet rationale under Stop Directive to a single sentence; the directive itself ('Halt the current thread. Reflection is a pause, not a side-quest') already conveys the rule.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core framework is tight (bias table, closing-recommendation table, error-handling table), but there is noticeable duplication: the 11-phrase trigger list repeats the description verbatim, the Anti-Patterns section restates rules already given under Tone and Format Rules and Error Handling, and the Stop Directive's 'This matters because' bullets explain rationale Claude could infer. More than 'minor instances' (4) but not pervasive padding (2). | 3 / 5 |
Actionability | For an instruction-only skill this is fully actionable: a verbatim optional-question template with numbered options and a default, exact output format strings ('Continue. {specific reasoning for why}.'), a bias recognition-cue table keyed to specific conversation patterns, and concrete error-handling mappings. The three cited scripts exist and parse as valid, stdlib-only executable Python with usage docs. | 5 / 5 |
Workflow Clarity | Clear execution sequence: halt the thread → conditional Q1 with an explicit stop condition ('One question max') → re-read the full conversation and run the five dimensions in order → mandatory closing with three specified outcomes. The error-handling table provides feedback paths for edge cases, and the skill involves no destructive or batch operations, so no validation cap applies. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview with detailed material correctly split into three real, one-level-deep reference files (cognitive_bias_canon, honest_output_discipline, conversation_reflection_practice — all verified present, cross-links are flat sibling links, no nesting), each clearly signaled with markdown links and one-line descriptions. Scripts are listed with roles in a Tooling table. Easy navigation throughout. | 5 / 5 |
Total | 18 / 20 Passed |