Content
88%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-structured routing skill: the body stays a concise dispatcher with explicit routing rules, validation checkpoints, and strong error-recovery guidance, while the heavy YAML detail lives in real one-level-deep reference files. The main weaknesses are a scope-playbook link that escapes the bundle and does not resolve, and duplicated boilerplate repeated across every reference file.
Suggestions
Fix the scope-establishment.md reference: the relative path ../../../references/scope-establishment.md escapes the skill directory and does not resolve — either inline the playbook into the skill's own references/ directory or reduce the Step 1 bullets to a single pointer to it.
Deduplicate the MANDATORY — Tool Result Verification, Scope Rules, and Experiment Components blocks that are repeated verbatim in create.md, edit.md, and components.md; keep one canonical copy (e.g., in components.md) and reference it from the other files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a lean ~60-line router with no concept-explainer padding, but the four Step 1 bullets ('Parse org / project from the user's message...', 'Ask only for what is missing...', 'Restate the active scope...', 'Never assume a default org/project...') restate rules the linked scope playbook already provides, and MANDATORY bullets 1 and 3 ('NEVER claim a tool call succeeded without an explicit success response' vs 'You MUST wait for the follow-up tool response confirming execution') overlap. | 4 / 5 |
Actionability | Guidance is fully executable for an instruction/routing skill: exact tool calls with parameters ('harness_execute(resource_type="chaos_experiment", action="run", resource_id="<experiment_id>")', 'harness_describe(resource_type="chaos_experiment")'), a verbatim clarifying question to ask when unsure, and concrete example utterances mapped to specific files. | 5 / 5 |
Workflow Clarity | Clear sequence (Step 1: Establish scope -> Step 2: Route before acting -> route-specific file) with explicit validation checkpoints ('Validate the manifest against the YAML Validation Rules in references/components.md before presenting it or saving', 'Wait for the tool's confirmed save response before reporting success') and Troubleshooting feedback loops (describe schema -> fix field -> retry; correct infrastructure filter -> re-run harness_list). | 5 / 5 |
Progressive Disclosure | The body appropriately routes to three one-level-deep reference files (components.md, create.md, edit.md) that all exist and are clearly signaled per intent, but the 'references/scope-establishment.md' link resolves outside the skill bundle (../../../references/scope-establishment.md is not present), and the MANDATORY, Scope Rules, and Experiment Components blocks are duplicated verbatim across all three reference files. | 4 / 5 |
Total | 18 / 20 Passed |