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.
An excellent, dense operational reference: fully executable curl recipes for every common case, a validated multi-step workflow, and thorough pitfall/error-code coverage. Its two weaknesses are repeated guidance across sections (rate limits, encoding, pretty-printing each appear 2-3 times) and the absence of any progressive-disclosure file split for the ~230 lines of inline API reference.
Suggestions
Split the field-types table and detailed query/sort/view recipes into a references/api.md and keep SKILL.md as an overview with the workflow and base patterns — the body is ~230 lines of inline API reference with no bundle files.
De-duplicate guidance stated in multiple places (rate limit in API Basics vs. Mutations vs. Pitfalls; filterByFormula encoding in Queries vs. Pitfalls vs. Notes; pretty-printing in API Basics vs. Notes) into single authoritative sections.
The pagination while-loop would be slightly stronger with a 429/Retry-After backoff inside the loop, since it makes the most requests of any pattern shown.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence — no 'what is Airtable/curl' padding — and every command is dense with signal, matching 'efficient; minor instances of over-explanation'. Not 5 because rate limits, filterByFormula encoding, and pretty-printing guidance are each repeated across API Basics, Mutations, Pitfalls, and 'Important Notes', which could be consolidated. | 4 / 5 |
Actionability | Every section is a copy-paste-ready curl command with real flags and headers: listing, filtering (with a working URL-encoding one-liner), sorting, creating, PATCH, upsert with performUpsert, delete, and a complete pagination while-loop. Matches the 'fully executable; specific examples cover the common cases' anchor; there is no vagueness or pseudocode anywhere. | 5 / 5 |
Workflow Clarity | 'Typical Hermes Workflow' is a clear 6-step sequence with explicit validation checkpoints (auth probe expecting 200, schema inspection before mutation, read-before-write to resolve record IDs) and an error-recovery feedback loop ('Read the errors array on non-2xx responses' plus 429/Retry-After handling). Destructive deletes get a confirm-before-firing step, so the batch/destructive cap at 3 does not apply; fits the anchor 'clear sequence with explicit validation steps; feedback loops for error recovery'. | 5 / 5 |
Progressive Disclosure | Headers are clear and navigation is easy, but this is a 229-line single file with no references/ bundle — the field-types table and the full query/mutation reference are inline material that could live in a separate file, matching 'some structure but... content that should be separate is inline'. Not 4/5 because the under-50-lines no-external-references exception does not apply at this length, and not 2 because the inline content is genuinely well-sectioned rather than a monolithic reference dump. | 3 / 5 |
Total | 17 / 20 Passed |