Content
50%Scale 1-3Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is a complex orchestration/routing skill for a multi-stage engineering workflow. It demonstrates good structural intent with clear sections for when to use/not use, procedure, validation, safety, and handoff routing. However, it suffers from abstract, jargon-heavy language without concrete executable examples, truncated procedures that depend on unavailable bundle files, and a cluttered references section that makes navigation difficult.
Suggestions
Add a concrete structured output example showing the actual JSON schema with realistic field values (schema_version, mode, stage_map, etc.) so Claude knows exactly what to produce.
Replace the abstract procedure steps with specific commands or tool invocations — e.g., show how to read .harness/session-evidence/latest.md, how to compare it against a plan file, and what the decision logic looks like in practice.
Consolidate the scattered 'See references/hot-path-folded-context.md' callouts into a single 'Folded Detail' section or inline the critical path content, since the bundle files are not available to verify these references.
Organize the References section into categorized groups (e.g., 'Core contracts', 'Situational references', 'Shared HE references') with one-line descriptions of when each is needed.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is moderately efficient but includes significant jargon-heavy prose that could be tightened. Phrases like 'Coordinate state, not ceremony' and repeated references to folded context files add overhead. Several sections (Safety Boundaries, Failure Handling) repeat concepts that could be consolidated, and the dense enumeration of output fields and reference paths is verbose without being maximally useful. | 2 / 3 |
Actionability | The skill provides structured procedural steps and clear routing rules, but lacks any concrete executable code, commands, or copy-paste-ready examples. The procedure is described at a high conceptual level ('Reconstruct lifecycle state from live repo, tracker, PR...') without showing exactly how to do it. The examples section describes scenarios in prose rather than showing actual inputs/outputs or structured output examples. | 2 / 3 |
Workflow Clarity | The procedure section has numbered steps and the validation section mentions pass/fail/blocked gates, which is good. However, the workflow lacks explicit validation checkpoints between steps, and the procedure steps are truncated with 'See references/hot-path-folded-context.md' without the bundle file being available. The feedback loop for error recovery is mentioned in Failure Handling but not integrated into the procedure sequence itself. | 2 / 3 |
Progressive Disclosure | The skill references multiple external files (references/contract.yaml, references/evals.yaml, references/hot-path-folded-context.md, several ../../references/ paths) which shows intent for progressive disclosure. However, no bundle files are provided to verify these exist, several references are deeply nested (../../references/), and the References section itself is a dense wall of links without clear categorization of when each is needed. The repeated 'See references/hot-path-folded-context.md for folded X detail' pattern across multiple sections is reasonable but the overall reference organization is cluttered. | 2 / 3 |
Total | 8 / 12 Passed |