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.
A highly actionable, well-validated operational skill: exact operation names, executable JS/JSON idioms, and explicit checkpoints throughout. Its main weakness is structure — half the file is inlined example-payload data that belongs in a references/ file, plus minor duplication between the workflow and gotchas sections.
Suggestions
Move the few-shot ui.playground.prompt.tools.write example payloads (the '### Few-shot examples' section, ~170 lines) into a references/ file such as references/tools-write-examples.md, keeping only the read-then-write revision idiom and a one-line pointer inline.
Deduplicate rules stated in both the workflow steps and the 'Things to avoid' section (raw-tool write restrictions, forced tool-choice reset, instance.add vs clone semantics) — state each once and cross-reference it from the other location.
Trim explanations of known concepts, e.g. 'LLM outputs are nondeterministic; repetitions build confidence by checking the same task across multiple runs' can be reduced to the operational rule about when to call ui.playground.repetitions.set.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational with almost no padding, but a few sentences explain concepts Claude already knows ("LLM outputs are nondeterministic; repetitions build confidence...") and several rules (instance.add vs clone, raw-tool write restrictions, forced tool-choice reset) are stated twice across the workflow and 'Things to avoid' sections. Minor trims are possible, matching the 'efficient; minor instances of over-explanation' anchor rather than the every-token-earns-its-place anchor. | 4 / 5 |
Actionability | Every workflow names exact operations (ui.playground.prompt.tools.read/write, ui.playground.run, etc.) with copy-paste-ready idioms including error checks (if (!run.ok) return run;), complete JSON example payloads for tools.write batches, a full phoenix-gql query for experiment results, and explicit defaults (snake_case parameter names, object/properties/required schema shape). This matches the fully-executable, common-cases-covered anchor. | 5 / 5 |
Workflow Clarity | Three numbered workflows are clearly sequenced with explicit validation gates (check run.ok, trust summaries only when job.status is COMPLETED or runCount == expectedRunCount), feedback loops (read revision before write, re-read between writes, edit-run-inspect-repeat until useful), and a gotchas checklist. The batch tools.write operation includes validation (fresh expectedRevision, all-or-nothing semantics with explained errors), so the batch-operation cap does not apply; this matches the anchor with explicit validation steps, error-recovery loops, and checklists. | 5 / 5 |
Progressive Disclosure | Sections are well organized (three workflows, few-shot examples, things to avoid) but the skill is entirely monolithic: roughly 170 lines of few-shot tools.write JSON examples are inlined in SKILL.md with no bundle files to absorb that detail. This matches the 'content that should be separate is inline' anchor rather than anchor 4, where most content is appropriately placed in separate files. | 3 / 5 |
Total | 17 / 20 Passed |