v1.2.1

forge-override: user-confirmed bypass for security hooks

Security hooks (command-guard, db-guard) now support an override mechanism for when a blocked command is intentionally needed.

How it works

1. A hook blocks a destructive command
2. Claude explains what was blocked and why
3. You confirm you want to proceed
4. Claude retries with # forge-override: <reason> as the first line
5. Claude Code's permission prompt shows the full command — you approve or deny
6. The override is logged to ~/.claude/security.log

What's new

• Override token — # forge-override: <reason> bypasses command-guard and db-guard
• Reason required — bare token or whitespace-only is rejected
• Audit trail — all overrides logged with timestamp, reason, and command
• Behavioural rules — new security-overrides.md constrains Claude from using overrides without your explicit confirmation
• SECURITY.md — full documentation of the override mechanism, security model, and what cannot be overridden

What cannot be overridden

• secret-filter — advisory only, fix false positives via better patterns
• architect-gate — plan quality enforcement
• commit-validator — AI attribution block

Stats

• 461 tests (10 new override tests)
• 8 rules files (was 7)

Fixes

• Corrected changelog dates for v1.0.0–v1.2.0 (were incorrectly listed as 2025)