Summary

Tracking issue for the current code-scanning backlog in NVIDIA/NemoClaw/security/code-scanning.

Note: this is a prioritization tracker for maintainers. It intentionally avoids exploit/reproducer detail; alert links require code-scanning access.

Refresh status — June 6, 2026

Last checked against:

• origin/main @ 7e0772448a388104d4ba2e01436e5f49980b8f65
• Latest completed Security / Code Scanning run on main: 27052564135, success, created 2026-06-06 04:28 UTC
  • CodeQL (javascript-typescript): success
  • CodeQL (python): success
  • ShellCheck SARIF: success
• Live code-scanning API inventory: GET /repos/NVIDIA/NemoClaw/code-scanning/alerts?state=open, paginated, with code-scanning access

Current open inventory:

What changed since the June 5 tracker refresh

• fix(security): close P1 code scanning findings #4880 merged the P1 cleanup for the old production logging/runtime correctness bucket and production-ish ShellCheck findings.
• fix(security): clean P2 tooling scan findings #4884 merged low-risk P2 cleanup for release-tooling docs URL regexes, test URL/host regex assertions, and remaining E2E ShellCheck findings.
• fix(security): harden P0 code scanning boundaries #4878 merged the P0 production hardening pass for runner/uninstall process boundaries, HTTP/file probes, and atomic/state/credential IO paths.
• refactor(security): extract curl probe policy #4885 merged a follow-up curl probe policy refactor and is included in the latest completed main scan above.
• The live inventory is now 120 open CodeQL alerts and 0 open ShellCheck alerts. This is a net reduction of 33 total alerts from the June 5 active inventory.
• The P0/P1/P2 merge train closed the previous ShellCheck backlog, the js/regex/missing-regexp-anchor bucket, the js/call-to-non-callable bucket, and most production process/file-system race findings.
• The latest inventory also surfaced 9 new js/clear-text-logging findings in production inference/onboarding paths (alert IDs 765-773). Treat these as the highest-signal next cleanup before broad test/docs quality work.
• Remaining production/runtime boundary stragglers are small: src/lib/inference/ollama/proxy.ts (js/file-access-to-http, alert 601) and src/lib/actions/uninstall/run-plan.ts (js/indirect-command-line-injection, alert 180). The rest of the command/env/temp-file/URL/TLS findings are test/tooling scoped.

Active CodeQL inventory by rule

Active ShellCheck inventory

Updated risk/reward plan

Proposed next PR sequence

1. Credential/logging cleanup PR: close new js/clear-text-logging alerts 765-773 in production inference/onboarding paths with redaction tests.
2. Boundary straggler PR: close or explicitly dismiss the two remaining production/runtime boundary alerts (src/lib/inference/ollama/proxy.ts and src/lib/actions/uninstall/run-plan.ts).
3. Advisor tooling boundary PR: close/dismiss tools/advisors/* and tools/e2e-advisor/* network/file artifact findings 594-596.
4. Test/E2E pattern cleanup PR(s): batch remaining test-only command/env, temp-file, URL/TLS, regex-character, and file-race alerts.
5. Static-quality cleanup train: docs extension cyclic imports/search asset plus JS/Python quality-only findings.

After each merge, wait for the next main Security / Code Scanning run, update this issue with the new total and rule deltas, and remove closed alert IDs from the active inventory.

Definition of done

• Live CodeQL alert inventory has been refreshed with authenticated current counts.
• P0 production CodeQL buckets are fixed or dismissed with explicit rationale.
• All 15 ShellCheck findings are fixed or suppressed with rationale and confirmed closed by the latest main scan.
• Remaining test/docs/tooling CodeQL findings are fixed, dismissed with rationale, or split into targeted follow-up issues.
• Security-sensitive production fixes have regression coverage.
• A final main code-scanning run confirms the remaining expected alert closures.