Content
76%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.
This is a deliberately minimal, auto-generated wrapper skill that fully delegates to a shared skill file outside the bundle. It is maximally concise and its single instruction is concrete and unambiguous, but it offers no visible workflow, validation, or task substance of its own, and the referenced path is unverifiable from the skill directory. It scores well on efficiency and structure while the substantive dimensions depend entirely on content that is not present here.
Suggestions
Since the skill performs autonomous cleanup of configuration (a batch, potentially destructive operation), add a brief validation/verification expectation at this level (e.g., 'Before applying cleanup changes, confirm the target config parses and back up modified files') so the workflow is not capped by the missing-checkpoint rule.
Add one line summarizing what the shared skill covers (e.g., which config files or friction sources it reviews) so the wrapper serves as a genuine overview even if the referenced path is unavailable.
Make the delegation resilient: note what to do if `~/.agents/skills/retro/SKILL.md` is missing (e.g., fall back to reviewing recent session friction points directly), since the current body dead-ends if the shared file is absent.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is two lines — a generation comment ('Generated by scripts/agent-harnesses.py; do not edit directly.') and a single delegation instruction ('Load and follow the shared skill at ~/.agents/skills/retro/SKILL.md.') — with zero padding and no explanation of concepts Claude already knows. Every token earns its place, matching the 'lean and efficient' anchor. | 5 / 5 |
Actionability | 'Load and follow the shared skill at `~/.agents/skills/retro/SKILL.md`' is a concrete, executable instruction with an exact path, but the body itself carries no task-level guidance or fallback — all substance is delegated to an external file. It is mostly executable guidance with a coverage gap within the document, fitting the 4 anchor rather than 5's 'specific examples cover the common cases'. | 4 / 5 |
Workflow Clarity | The single delegation step is unambiguous, which would normally earn 5 under the simple-skill exception, but the skill's stated purpose is 'autonomous cleanup' of agent/Codex/Claude configuration — a batch, modification-heavy operation — and the body shows no validation or verification checkpoint. The rubric explicitly caps destructive/batch skills without validation at 3, taking precedence over the simple-skill exception. | 3 / 5 |
Progressive Disclosure | The body is a clean wrapper with one clearly-signaled, one-level-deep reference — trivially navigable. It falls short of 5 because the reference points outside the skill bundle to an absolute home-directory path that does not resolve in this environment, and the wrapper includes no overview of what the delegated skill actually covers. | 4 / 5 |
Total | 16 / 20 Passed |