Content
75%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 strong, information-dense skill body: executable SQL and TypeScript throughout, precise Paddle status semantics, and a genuinely useful pitfalls section. The main gaps are minor — an incomplete placeholder in the backfill script, no verification checkpoint for the batch backfill, and MCP tool semantics inlined in the main file.
Suggestions
Complete the backfill script's `update: { /* same fields */ }` placeholder with the actual field list and show (or link) the `get-paddle-instance` helper so the code is copy-paste ready.
Add a verification step for the batch backfill (e.g. compare row counts against the Paddle dashboard, or spot-check a known customer) before flipping access gating to read from the database.
Move the embedded MCP tool-semantics blockquote (50-call cap, pagination shape, casing rules) into a `references/paddle-mcp.md` file and link to it, slimming the main SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious domain knowledge (service-role key/RLS gotcha, `canceled` as terminal state, the 50-call MCP cap, out-of-order event handling) with almost no padding; minor prose like "Why mirror state at all?" could be trimmed, keeping it just below the lean level-5 anchor. | 4 / 5 |
Actionability | The SQL schema, TypeScript webhook handlers, access-gate and UI-state functions are concrete and executable, and both backfill paths are shown; the backfill script ships a literal `update: { /* same fields */ }` placeholder and relies on an unshown `get-paddle-instance` helper, so it is not quite copy-paste ready per the level-5 anchor. | 4 / 5 |
Workflow Clarity | A clear sequence (prerequisites, schema, handlers, status semantics, scheduled changes, queries, backfill, pitfalls) plus an explicit "Verify the integration" section with concrete checks; the backfill is a batch operation with no verification checkpoint of its own and no fix-and-retry loop, which the rubric's batch-operation guidance counts as a validation gap capping it below 5. | 4 / 5 |
Progressive Disclosure | A well-sectioned single file with clearly signaled external links and no nested references; the embedded MCP tool-semantics blockquote and the ~330-line monolith are content a separate reference file could absorb, so it falls between the well-placed level 4 and the cleanly split level 5. | 4 / 5 |
Total | 16 / 20 Passed |