docs: restore security documentation as SECURITY.md

[derekmisler]

Restores the security documentation removed when bash scripts were replaced by TypeScript. The content has been audited and updated to reflect the current src/security/ implementation. Also fixes two broken links in README.md that pointed to the now-deleted security/README.md.

[docker-agent]

Assessment: 🟡 NEEDS ATTENTION

One documentation accuracy issue found in the newly added SECURITY.md. The README.md link fixes are correct. The security architecture description, authorization tier table, module table, CLI interface docs, pattern tables, and test coverage claims all accurately reflect the current src/security/ implementation.

[derekmisler]

Good catch — fixed. Changed "two" to "three" to match the actual implementation (and the Heuristic 3 (structural validator) label in the source).

[docker-agent]

Assessment: 🟢 APPROVE

This PR accurately restores security documentation as SECURITY.md. The added content was verified against the actual src/security/ TypeScript implementation:

• Authorization tiers — correctly describes the 4-tier waterfall in src/main/auth.ts (skip-auth → trusted-bot → org-membership → author_association)
• Output scanning heuristics — correctly states three false-positive heuristics for sanitizeOutput (metacharacter check, single-quote wrapping check, structural CRC32 validator). The source code's JSDoc comment says two but the implementation has three — the documentation is more accurate than the stale JSDoc
• Prompt sanitization — correctly describes the three-tier strategy (CRITICAL/SUSPICIOUS/MEDIUM-RISK) plus the diff-comment preprocessing step, matching src/security/sanitize-input.ts's own terminology
• CLI subcommand signatures — check-auth, sanitize-input, sanitize-output verified against src/security/index.ts
• SECRET_PATTERNS table — all 7 patterns (Anthropic, 5× GitHub, OpenAI) verified against src/security/patterns.ts
• SECRET_PREFIXES alternation string — including sk-proj- verified as present in the actual code
• GitHub Actions output names — secrets-detected, prompt-suspicious, input-risk-level, security-blocked verified against src/main/index.ts and action.yml
• Module file references — src/security/validators.ts, src/main/auth.ts, src/main/index.ts all confirmed to exist
• Test case counts — 21 cases from test-security.sh + 6 from test-exploits.sh confirmed in security.test.ts header