fix(prd-7): M3 security review — self-trigger guard, actor permissions, trigger bypass, platform validation

- Add self-mention/bot loop prevention in actor.ts (blocks actor matching bot_name)
- Switch Forgejo permission check to actor-level via getCollaboratorPermissionLevel
  with fallback to token-level repos.get() when collaborator API returns 403
- Close empty-body PR review comment trigger bypass (now returns false instead of true)
- Add platform detection cross-validation (reject PLATFORM=forgejo on github.com)
- Document URL builder defense-in-depth rationale in comment-logic.ts
- Update/add tests across 6 test files (actor, permissions, trigger, platform, mcp, m7)

Progress: 50% complete (3/6 milestones) — next: M4 test coverage review

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Vlad MocanuandClaude Opus 4.6 committed 2026-02-26 12:31:12 +02:00
1 parent cefc229199
commit 6b0302d226
12 files changed
+270 -68

No files matched your search

+6
View File
@@ -51,6 +51,12 @@ export const FORGEJO_SPINNER_HTML = "⏳";
// URL Builders
// ---------------------------------------------------------------------------
// Note: owner, repo, runId, and branchName are NOT URI-encoded in path
// segments because: (1) owner/repo are validated by Forgejo (alphanumeric +
// .-_), (2) branchName is validated by validateBranchName() which restricts
// to [a-zA-Z0-9._/-], and (3) encoding "/" in branch names breaks Forgejo
// URLs (e.g., "claude/fix-bug" → "claude%2Ffix-bug" is invalid).
export function createForgejoJobRunLink(
owner: string,
repo: string,