Skip to content

ref(skill-drift): scan references/sdks instead of legacy skills#266

Open
evanpurkhiser wants to merge 1 commit into
mainfrom
evanpurkhiser/ref-skill-drift-scan-references-sdks-instead-of-legacy-skills
Open

ref(skill-drift): scan references/sdks instead of legacy skills#266
evanpurkhiser wants to merge 1 commit into
mainfrom
evanpurkhiser/ref-skill-drift-scan-references-sdks-instead-of-legacy-skills

Conversation

@evanpurkhiser

Copy link
Copy Markdown
Member

The per-SDK skills the nightly drift automation used to keep fresh have moved to skills-legacy/ (#265) and are now frozen. The maintained platform content lives in the references/sdks/<sdk>/ trees, so this repoints the automation there.

  • Matrix (.github/skill-drift-matrix.json): the unit is now an SDK slug (sdk: python) rather than a skill name, matching the reference directory names. repo / path_filter / team are unchanged.
  • Detect job: the sandboxed Claude session reads references/sdks/<sdk>/ (index.md plus the per-signal files) and is told to conform edits to references/sdks/STRUCTURE.md. Its file-edit allowlist, the summary.json schema (.sdk instead of .skill), and the artifact staging all target that tree.
  • Apply job: the path allowlist, rsync overlay, branch/slug naming, and the prompt-injection guard (now a bare-slug check instead of the sentry-*-sdk pattern) all operate on references/sdks/<sdk>/.

The security model is unchanged — read-only detect job, deterministic apply job, path allowlists enforced in both.

Sequencing: this reads the checked-out tree at run time and only fires weekly (or on manual dispatch), so it activates once references/ lives at the repo root via the skills-next merge. Should land alongside that merge and #265.

The per-SDK skills the drift automation used to keep fresh have moved to
skills-legacy/ and are now frozen. The maintained platform content lives in the
references/sdks/<sdk>/ trees, so repoint the automation there.

- Matrix: the unit is now an SDK slug (`sdk: python`) rather than a skill name,
  matching the reference directory names.
- Detect: the Claude session reads references/sdks/<sdk>/ (index.md plus the
  per-signal files), conforms edits to references/sdks/STRUCTURE.md, and its
  file-edit allowlist, summary schema (.sdk), and artifact staging all target
  that tree.
- Apply: the path allowlist, overlay, branch/slug naming, and the prompt-
  injection guard (now a bare-slug check) all operate on references/sdks/<sdk>/.

Activates once references/ lives at the repo root (the skills-next merge); the
job reads the checked-out tree at run time and runs weekly.

@cursor cursor Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Cursor Bugbot has reviewed your changes and found 1 potential issue.

Fix All in Cursor

❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.

Reviewed by Cursor Bugbot for commit 3748d3e. Configure here.

```

If you see one titled `[skill-drift] ...<SKILL_NAME>...`, treat that
If you see one titled `[skill-drift] ...<SDK>...`, treat that

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Bare SDK slug false dedupe

Medium Severity

Step 1 tells the detect agent to treat any open [skill-drift] PR or issue whose title contains the bare SDK slug as already covered. Short slugs like go or react match as substrings in unrelated titles (e.g. legacy sentry-go-sdk titles or other words), so the run can emit no_drift and skip real reference updates.

Fix in Cursor Fix in Web

Reviewed by Cursor Bugbot for commit 3748d3e. Configure here.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants