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 high-quality operational skill: the workflows are unambiguous with explicit confirmation and verification checkpoints, error codes map to concrete recovery actions, and commands are copy-paste ready. The main improvements are trimming repetition (the CLI rule/name limitation is stated three times) and moving more of the inline operational detail into bundle reference files.
Suggestions
State the CLI limitation on updating `rule`/`name` once (e.g. in a single 'CLI gaps' note) instead of three times in Create Step 6, Update Step 2, and the Troubleshooting table.
Move the Troubleshooting table and the traffic-readiness field checks into a references/ file linked from the relevant steps, keeping SKILL.md closer to an overview-plus-workflow and reducing inline tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious operational detail (exact params, error-code semantics, CLI gaps) and assumes Claude's competence rather than explaining concepts, fitting the 4 anchor 'efficient; minor instances of over-explanation'. Not 5 because some content is repeated: the CLI limitation on changing `rule`/`name` appears in Create Step 6, Update Step 2, and the Troubleshooting table, and a few compound sentences could be tightened. | 4 / 5 |
Actionability | Fully executable guidance: exact MCP invocations ("harness_update · fme_experiment · params: { experiment_id } · body: { description?, hypothesis?, ... }"), copy-paste-ready CLI commands ("harness update experiment <experiment-id> --set description=foo --set status=PAUSED"), concrete constraints (name 2-250 chars, hypothesis max 500 chars, "Do not send assignmentSource (400 if present)"), and error-specific recovery actions. Matches the 5 anchor: specific commands cover the common cases. | 5 / 5 |
Workflow Clarity | The 8-step create flow is clearly sequenced with explicit validation checkpoints ("**STOP HERE.** Wait for explicit confirmation"; Step 8 compares every field against the approved draft and stops on mismatch) and error-recovery feedback loops (409 → ask for a new name, 404 → back to Step 1). The destructive delete flow requires stricter confirmation and offers a safer alternative, satisfying the destructive-operation validation requirement. | 5 / 5 |
Progressive Disclosure | Good structure: the one bundle file (references/stop-conditions.md) is real, well-signaled, contextually linked from the right steps, and one level deep; bulk detail (tool-map, write-safety, concepts) lives in separate repo-level reference files. Falls short of 5 because the SKILL.md itself is a long operational manual with some content (traffic-readiness field checks, troubleshooting rows duplicating in-body guidance) that could be split into a reference file. | 4 / 5 |
Total | 18 / 20 Passed |