fix(cursor): discover symlinked skills as package boundaries by EzraBuild · Pull Request #9420 · pingdotgg/t3code

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

EzraBuild · 2026-09-03T12:18:52Z

What Changed

Cursor skill discovery (CursorSkills.ts, added in #9180) canonicalizes each directory and skips any whose resolved path falls outside the scanned root. A skill installed as a symlink into a config repo, for example ~/.cursor/skills/my-skill -> ~/config-repo/agents/skills/my-skill, resolved outside the root and was silently dropped. The $ picker stayed empty and $my-skill was never rewritten to Cursor's native /my-skill.

This treats an escaping symlink as a package boundary: read that directory's own SKILL.md under the link name, but never walk the target tree. The visited set still prevents loops and the existing entry and byte budgets still bound the scan. Nothing outside the root is ever traversed.

One new test symlinks a skill from an outside directory into .cursor/skills, asserts it is discovered under the link path, asserts a nested SKILL.md below the target is not discovered, and asserts the probe reports success.

Why

Fixes the remaining case in #2736, as reported in the 2026-09-03 comment there. Claude's scanner already follows these symlinks and the Antigravity scanner ships a test for the same behavior, so Cursor was the odd one out on the same machine.

UI Changes

None.

Checklist

This PR is small and focused
I explained what changed and why
I included before/after screenshots for any UI changes (not applicable)
I included a video for animation/interaction changes (not applicable)

Validation

vp test run apps/server/src/provider/Layers/CursorProvider.test.ts
vp lint and vp fmt --check on the two changed files
vp run --filter t3 typecheck

Model: Claude Fable 5.1. Harness: Claude Code.

Cursor skill discovery canonicalized every directory and skipped any that resolved outside the scanned root. A skill installed as a symlink into a config repo therefore never reached the `$` picker or the `$skill` to `/skill` rewrite, while Claude and Codex on the same machine listed it. Treat an escaping symlink as a package boundary: read its own SKILL.md under the link name, but never walk the target tree. The visited set and scan budgets still bound the work. Fixes part of pingdotgg#2736. Model: Claude Fable 5.1. Harness: Claude Code.

macroscopeapp · 2026-09-03T12:24:25Z

Approvability

Verdict: Approved at 5e919d7

Macroscope's review found this PR approvable — This is a narrowly scoped Cursor skill-discovery bug fix that recognizes externally symlinked packages without traversing their target trees. Existing cycle and scan-budget protections remain in place, and a regression test covers both discovery and the traversal boundary.

^{You can add or adjust custom eligibility rules. Learn more.}

github-actionsBot added vouch:unvouched PR author is not yet trusted in the VOUCHED list. size:S 10-29 changed lines (additions + deletions). labels Sep 3, 2026

macroscopeappBot approved these changes Sep 3, 2026
View reviewed changes

Melv1C approved these changes Sep 3, 2026
View reviewed changes