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 dense, highly actionable reference skill: exact runner labels and pricing tables, executable YAML/CLI examples, correct/incorrect label syntax, and a strong error-string-to-fix troubleshooting table. The gaps are minor — duplicated Depot-CI disambiguation prose, absent post-setup verification steps, and a fully inline single-file structure with no reference files for the longer integration details.
Suggestions
Add a verification step to the Setup and Tailscale sequences (e.g., push a test workflow on a depot-ubuntu label and confirm a runner is assigned) to close the workflow_clarity checkpoint gap.
Collapse the 'Depot runners are not Depot CI' section to one sentence with a single link to the depot-ci skill — it currently makes the same point twice and links the same reference twice, trimming tokens for conciseness.
Move the per-tool cache table, Xcode caching details, and Tailscale ACL examples into a references/ file linked from short overview sections, so SKILL.md functions as the navigable overview progressive disclosure expects.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with product-specific data Claude cannot know (label/pricing tables, pre-configured cache env vars) and never explains basics like what GitHub Actions is; version-bearing label names are the product's API surface rather than incidental time-sensitive detail, and the Dagger section is properly quarantined under "**Deprecated:**". It falls short of anchor 5 only through minor trims: the "Depot runners are not Depot CI" section states its disambiguation twice and links ../depot-ci/references/posthog-check-run-semantics.md twice with overlapping framing, and the single-paragraph Egress Filtering section packs several rules into one block. | 4 / 5 |
Actionability | Everything is executable: a before/after migration YAML snippet ("# Before: runs-on: ubuntu-latest / # After: runs-on: depot-ubuntu-24.04-4"), copy-paste CLI ("depot org show", "depot org switch <org-id>"), exact wrong/right label syntax, a Tailscale ACL JSON example, and a troubleshooting table mapping exact error strings ("No space left on device", "Unable to get ACTIONS_ID_TOKEN_REQUEST_URL") to specific fixes including a concrete "POST /repos/{owner}/{repo}/actions/runs/{id}/force-cancel" call. This matches anchor 5's copy-paste-ready coverage of common cases. | 5 / 5 |
Workflow Clarity | Setup (3 numbered steps) and Tailscale (4 numbered steps) are clearly sequenced, and the Troubleshooting table supplies error-to-fix recovery paths that function as feedback loops. It sits at anchor 4 rather than 5 because no checkpoint validates success (e.g., nothing confirms a workflow actually picked up a Depot runner after Setup, or that the tailnet is reachable after Tailscale step 4); it is above anchor 3 because sequences are complete and recovery guidance for known failures is explicit, and no destructive/batch operation applies to invoke the cap. | 4 / 5 |
Progressive Disclosure | No bundle exists (references/, scripts/, assets/ are all absent), so all ~220 lines live in SKILL.md; structure is nonetheless good — one section per topic, external docs clearly signaled under "Reference docs:", and links only one level deep (depot.dev URLs and the cross-skill depot-ci reference). It does not reach anchor 5's 'content appropriately split' because everything is inline in one file: the per-tool cache table, Xcode caching details, and Tailscale ACL examples are candidates for a references/ file; it stays above anchor 3 since nothing is buried or nested and sections are easy to navigate. | 4 / 5 |
Total | 17 / 20 Passed |