chore(security): pin claude-code-action to known-good v1.0.174

[abir-vim]

Automated fleet sync of the org security scanner. Pins claude-code-action off the floating @v1 tag (which regressed at v1.0.175+ and silently no-ops on Bedrock, failing the scan gate closed) to the known-good SHA. See scanner-org for the source template.

[github-actions]

Security Scanner Results

Critical	High	Medium	Low	Info	Dep BLOCK	Dep WARN
0	0	0	1	1	0	0

Findings

• 🔵 Security scanner workflow now allows all bot-authored PRs to trigger it (allowed_bots: "*") — .github/workflows/security.yml:170
  • Remediation: Prefer an explicit allowlist (e.g. allowed_bots: "renovate[bot],dependabot[bot]") over "*" so only the intended automation can trigger the scanner. If a wildcard is intentional for org-wide reuse, keep the accompanying comment documenting the private-repo + fixed-prompt assumptions so the trust model stays visible.
• ⚪ Scanner LLM reads attacker-influenced PR diff/body via git diff and gh pr view (prompt-injection surface) — .github/workflows/security.yml:174
  • Remediation: No change required for this PR. Keep the model's tool allowlist read-only and its output confined to findings.json (already the case), and continue rendering PR comments deterministically rather than letting the model post. Optionally add a note in the prompt reminding the model that diff/PR-body text is untrusted data, not instructions.

Gate blocks at severity critical or a dependency marked BLOCK. Full history in scanner-stats.