Content
78%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.
The body is an unusually dense and lean operational reference: exact model IDs, per-host parameter rules, validation checkpoints, and paid-job recovery guidance with essentially zero padding. Its main weaknesses are that all per-provider detail lives inline in SKILL.md with no reference-file split, and no complete worked request example is provided.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every line states a provider-specific fact Claude could not know, e.g. "fal v4 accepts at most 5,000 characters and has no speed/style control. Direct v4 accepts up to 10,000." and "LTX-2.5 supports text/image/audio generation, not retake or extend." There is no tutorializing, no explanation of known concepts, and no padding to trim, matching the lean anchor rather than the score 4 "minor over-explanation" anchor. | 5 / 5 |
Actionability | Concrete specifics throughout: exact IDs ("gpt-image-2.5-flare", "kling-o3-standard", "eleven_v4"), parameter names ("image_path", "mask_path", "resume_job"), env requirements ("use FAL_KEY, not the official Kling credentials"), and numeric limits ("up to 14 reference images", "at most 2,000 UTF-16 code units"). Score 5 is not reached because parameters and values are never assembled into one complete worked request example covering a common case end to end. | 4 / 5 |
Workflow Clarity | Sequenced steps and checkpoints are present: "First `list_looks` or `inspect_look`... before POST", "checked-in schemas under tools/provider_contracts validate the complete request before submission", "`dry_run=true` validates and prices the exact request without generating", and the timeout recovery loop "pass the returned `resume_job` to the same tool... check the provider dashboard before submitting again". Score 5 is not reached because the core route-validate-submit-recover flow is never laid out as a single ordered sequence; checkpoints are scattered across sections. | 4 / 5 |
Progressive Disclosure | Section headers are clear (Routing, Image requests, Video and jobs, Speech and music, Avatar V) and the one true reference is a well-signaled link to [provider-refresh-oct23], but roughly 120 lines of per-provider contract detail are inlined in SKILL.md that would navigate better as per-provider reference files, and other pointers (AGENT_GUIDE.md, docs/provider-update-plan-2026-10-03.md, tools/provider_contracts) are prose mentions rather than clearly signaled references. This matches the score 3 anchor of structure present but content that should be separate is inline; not 4 because the bulk detail belongs at reference depth, not 2 because organization is genuinely good. | 3 / 5 |
Total | 16 / 20 Passed |