Content
82%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 well-crafted workflow/policy skill: highly concrete and workspace-specific, with exact commands, paths, and env vars, and a strong verify-before-reporting checkpoint guarding the main failure mode. The multi-step sequence must be inferred from topic order rather than an explicit ordered flow, and all policy content is inlined in a single file with detail delegated to sibling skills rather than bundle references.
Suggestions
Add a brief numbered workflow overview (classify the ask → scaffold with `agent-native add-app` → verify on the gateway → build out actions/UI/data → finish branding) so the execution sequence is explicit; the current topic-ordered sections leave the order to be inferred and place scaffolding/verification after the data and DB policy sections.
Trim the redundant restatement in Scaffolding — "Never report the app as created while the preview still shows something else" repeats the preceding confirm-before-reporting rule — and compress optional rationale clauses ("so the app hydrates correctly…", "since the preview root can still show a different app") to save tokens.
If the skill grows, move self-contained policy domains (Vault Access, PostgreSQL Database, discovery/manifest configuration) into one-level-deep reference files under references/ so SKILL.md stays a lean overview with clearly signaled navigation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a dense rule-set of workspace-specific facts Claude cannot know (vault access modes, `AGENT_NATIVE_WORKSPACE_APPS_JSON`, `configureClientRouterBasename()` placement, `/_agent-native/actions/:name`) with no padding about already-known concepts. A few spots could be trimmed — "Never report the app as created while the preview still shows something else" restates the preceding confirm-before-reporting instruction, and rationale clauses like "so the app hydrates correctly when mounted at /<app-name>" are optional — matching "efficient; minor instances of over-explanation" rather than the every-token-earns-its-place top anchor. | 4 / 5 |
Actionability | Gives copy-paste-ready specifics throughout: the exact scaffold command "`pnpm exec agent-native add-app <app-name> --template=<template>` from the workspace root", env var names, file-level detail ("`configureClientRouterBasename()` call in `app/entry.client.tsx`, before `hydrateRoot`"), hook names (`useActionQuery`/`useActionMutation`), and an executable verification step ("request `/<app-name>` on the running gateway and confirm it returns the new app"). Implementation depth is appropriately delegated to named sibling skills, matching the fully-executable anchor for an instruction-only skill. | 5 / 5 |
Workflow Clarity | The Scaffolding section contains an explicit validation checkpoint with error recovery ("Before saying the app is created, request `/<app-name>` on the running gateway and confirm it returns the new app, not a fallback route", plus the misconfigured-host remediation naming where to set the dev command). However, the overall lifecycle (classify → scaffold → verify → build → finish) is never enumerated and must be inferred from topic-ordered sections, with scaffolding/verification appearing after the data and database policy sections — fitting "clear sequence with most checkpoints present; minor validation gaps" rather than the enumerated-checklist top anchor. | 4 / 5 |
Progressive Disclosure | Nine well-headed sections plus a clearly signaled, one-level-deep "Related Skills" list and an inline pointer to `.agents/skills/shadcn-ui/SKILL.md` make navigation easy, and no referenced path is missing (the skill has no bundle directories). But the ~110-line body inlines several full policy domains (Vault Access, PostgreSQL Database, discovery/manifest configuration) with no internal reference split — good structure with minor organization gaps rather than the appropriately-split top anchor, and the under-50-line no-references exception does not apply. | 4 / 5 |
Total | 17 / 20 Passed |