[sighthound] Security findings in github/gh-aw

[github-actions]

Sighthound scan flagged 116 actionable findings; top 5 by severity (all Critical/High confidence, "Command Injection" pattern) below.

1. pkg/cli/runner_guard.go:114 — exec.Command(dockerPath, "run", "--rm", "-v", volumeMount, ...). volumeMount appears built from dynamic/untrusted path data. Remediation: validate/sanitize the mount path, avoid shell interpolation, keep args as a fixed slice (already the case) but confirm volumeMount and containerScanPath are derived only from trusted, validated local paths.

2. pkg/cli/grant.go:176 — exec.Command(dockerPath, "run", "--rm", "-v", volumeMount, GrantImage, "--config", containerPolicyPath, ..., imageRef). imageRef/volumeMount may originate from user/workflow input. Remediation: validate imageRef against an expected image-name pattern before use; ensure paths are canonicalized and confined to expected directories.

3. pkg/cli/poutine.go:229 — exec.Command(dockerPath, "run", "--rm", "-v", volumeMount, "-w", "/workdir", "ghcr.io/boostsecurityio/poutine:latest", "analyze_local", ".", ...). Remediation: verify volumeMount is constructed from a sanitized, absolute repo path with no attacker-controlled segments.

4. pkg/cli/poutine.go:109 — Same pattern/image as above (duplicate call site). Remediation: same as Add workflow: githubnext/agentics/weekly-research #3 — validate volumeMount construction.

5. pkg/cli/upgrade_command.go:502 — exec.Command(exe, newArgs...). exe/newArgs may include dynamic values (e.g., from CLI re-exec/upgrade flow). Remediation: confirm exe resolves to a known trusted binary path (not attacker-influenced) and newArgs values are validated/allow-listed rather than passed through unchecked.

General guidance for all: avoid sh -c/shell string concatenation (already avoided via slice args), but ensure every dynamic value feeding into exec.Command arguments (paths, image refs, flags) is validated/allow-listed at the point of construction, since Sighthound flags these as High-confidence Command Injection risks due to untrusted data reaching exec.Command.

See /tmp/gh-aw/agent/sighthound/actionable.json for full details (148 total findings, 116 actionable outside test/testdata).

Generated by 🛡️ Sighthound Security Scan · auto · 17.3 AIC · ⌖ 3.26 AIC · ⊞ 7.6K · ◷

[github-actions]

🍪 Issue Monster selected this for Copilot

I have identified this issue as a good candidate for automated resolution and requested assignment to the Copilot coding agent.

If assignment succeeds, the Copilot coding agent will analyze the issue and create a pull request with the fix.

Om nom nom! 🍪

🍪 Om nom nom by Issue Monster · gpt54 · 3.05 AIC · ⌖ 5.56 AIC · ⊞ 8.4K · ◷