[bercedev]

Closes #3687

The Claude Code and Cursor hooks run chainloop trace hook … directly, so in a repository with the hooks committed but without the CLI on PATH, every hook fails with exit 127 before Chainloop's best-effort handling can run. This wraps them in the guard from the issue, the same one #3684 added for Codex:

sh -c 'if ! command -v chainloop >/dev/null 2>&1; then exit 0; fi; exec chainloop trace hook claude post-tool-use'

• A new trace.GuardHookCommand(command) builds that string. Both providers use it, so the guard is defined in one place.
• Claude Code (.claude/settings.json): buildHookEntry writes the guarded command. addHookEntry now treats an entry as up to date only if both its matcher and its Chainloop command match. Previously it compared only the matcher, so an unguarded entry from an earlier trace init would never have been upgraded. A small chainloopHookCommand helper finds the Chainloop handler in a group.
• Cursor (.cursor/hooks.json): buildHookEntry writes the guarded command, and addHookEntry compares against it. Cursor's flat entry format is unchanged.
• Unchanged as the issue asks: user hooks are untouched, trace init stays idempotent, and the Git hooks and the pre-push requireTrace enforcement are untouched. exec passes stdin and stdout through and keeps the CLI's exit status when it's present.

Tests, in both claude/hooks_test.go and cursor/hooks_test.go:

• TestInstalledHooksWithoutCLI runs every hook command with PATH holding only sh. Each one exits 0 with no output, and every installed command contains the guard.
• TestInstalledHookPreservesIOAndExitStatus runs against a stub chainloop that echoes its arguments and stdin and exits with 0 or 37. The output and exit code come through unchanged.
• TestInstallUpgradesUnguardedHooks starts from a settings file with a bare chainloop trace hook … post-tool-use entry next to a user hook. InstallHooks replaces it with exactly one guarded entry, keeps the user hook, and a second run changes nothing.
• The existing command assertions now expect the guarded form.

go test ./app/cli/internal/trace/..., go vet and gofmt -l are clean (Go 1.27.3, offline). Against main's hooks.go the new install and upgrade tests fail. The runtime tests skip on Windows, as Codex's do.

One question: Codex also sets a commandWindows with a cmd.exe/where guard. As far as I know Claude Code needs Git Bash on Windows, so sh should be available there, but I couldn't check it on Windows. I don't know of an equivalent override field for Cursor, so Cursor on Windows needs sh on PATH. If that's a concern, I can make the Cursor guard platform-specific at install time.

[chainloop-platform]

PR validation — ✅ 3 passing

Status	Policy	Material	Messages
✅ Passed	pr-min-approvals	pr-info	-
✅ Passed	pr-description-required	pr-info	-
✅ Passed	pr-user-story-linked	pr-info	-

_{View attestation ↗}

AI Session Checks — ⏭️ bypassed by label

AI Coding Session Check Bypassed

This PR carries the skip-ai-session label, so the AI coding session check was bypassed.

Learn more about Chainloop Trace.

Security Checks — ✅ 5 passing

✅ secret-scan

Status	Policy	Messages
✅ Passed	secrets-detection	-

✅ sast-scan

Status	Policy	Messages
✅ Passed	owasp-top10-2025	-
✅ Passed	sast	-
✅ Passed	cwe-top25	-
✅ Passed	cwe-top26-40-cusp	-

Scans not applied (3)

Scan	Reason
vulnerability-scan	no manifest/lockfile changed
github-actions-scan	no workflow files changed
iac-scan	no IaC files changed

_{View attestation ↗}

Security context

✅ Nothing this change touches has a recorded security-fix history.

_{View in Chainloop ↗ · How this works ↗}

---

Powered by Chainloop and Chainloop Trace

[cubic-dev-ai]

1 issue found across 5 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. When an issue isn't valid or won't be fixed in this PR, reply in its thread with the reason and then resolve the thread. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="app/cli/internal/trace/provider.go">

<violation number="1" location="app/cli/internal/trace/provider.go:404">
P2: The shared wrapper hard-codes POSIX `sh`, so Claude and Cursor hooks fail on Windows installations without Git Bash instead of skipping cleanly. Generate a Windows-compatible guard alongside the POSIX command, or use each provider’s platform-specific command fields.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

View guided diff | Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 6 files (changes from recent commits).

<sub>Reply with feedback, questions, or to request a fix.

View guided diff | Re-trigger cubic</sub>

[jiparis]

Thanks for your contribution, @bercedev. I'm merging this.

[jiparis]

Sorry @bercedev , none of the commits are signed. Please sign them before we can merge it.

[bercedev]

signed now, all four commits show as verified on github. same content as before (no changes beyond the signatures, rebased on the same base).

[jiparis]

Merging. Thanks @bercedev !