fix(prd-7): M4 test coverage review — 28 new tests for comment operations and branch setup

- test/create-initial-comment.test.ts: 15 tests covering GitHub/Forgejo
  platform branches, sticky comment matching, review comment fallback,
  error handling with GITHUB_OUTPUT verification
- test/update-tracking-comment.test.ts: 11 tests covering body construction,
  branch link logic (issue vs PR), comment type routing, Forgejo URLs
- test/setup-branch-forgejo.test.ts: 2 tests verifying Forgejo uses
  repos.getBranch() while GitHub uses git.getRef() for source branch SHA
- test/COVERAGE-NOTES.md: documents all coverage status and accepted gaps
  (tag mode orchestrator, entrypoint) with rationale
- Notable finding: CLAUDE_APP_BOT_ID discrepancy flagged for M5

Progress: 67% complete (4/6 milestones) — M5 code quality next

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

test/COVERAGE-NOTES.md

@@ -0,0 +1,51 @@
+# Test Coverage Notes — Forgejo Platform Support
+
+## Well Covered (unit tested)
+
+| Module                           | Tests           | File                                 |
+| -------------------------------- | --------------- | ------------------------------------ |
+| `src/forgejo/token.ts`           | 8 tests         | `test/forgejo-token.test.ts`         |
+| `src/platform/detector.ts`       | 17 tests        | `test/platform-detector.test.ts`     |
+| `src/forgejo/comment-logic.ts`   | 30+ tests       | `test/forgejo-comment-logic.test.ts` |
+| `src/forgejo/data/fetcher.ts`    | 50+ tests       | `test/forgejo-data-fetcher.test.ts`  |
+| `src/github/validation/actor.ts` | 6 Forgejo tests | `test/actor.test.ts`                 |
+| `src/mcp/install-mcp-server.ts`  | 8 Forgejo tests | `test/install-mcp-server.test.ts`    |
+
+## Added in M4
+
+| Module                                                 | Tests    | File                                   |
+| ------------------------------------------------------ | -------- | -------------------------------------- |
+| `src/github/operations/comments/create-initial.ts`     | 15 tests | `test/create-initial-comment.test.ts`  |
+| `src/github/operations/comments/update-with-branch.ts` | 11 tests | `test/update-tracking-comment.test.ts` |
+| `src/github/operations/branch.ts` (API divergence)     | 2 tests  | `test/setup-branch-forgejo.test.ts`    |
+
+## Accepted Gaps (orchestrators)
+
+### `src/modes/tag/index.ts` — `prepareTagMode()`
+
+**Why not unit tested**: Calls 8+ modules (checkHumanActor, createInitialComment,
+fetchForgejoData/fetchGitHubData, setupBranch, configureGitAuth, createPrompt,
+prepareMcpConfig). Each is individually unit tested.
+
+**Risk**: Low. The Forgejo branch is a simple `if (platform === "forgejo")`
+choosing `fetchForgejoData` vs `fetchGitHubData`. Both are thoroughly tested.
+
+**Coverage path**: Integration tests in the separate `install-test` repo.
+
+### `src/entrypoints/run.ts` — main orchestrator
+
+**Why not unit tested**: Token setup path is a trivial conditional (1 line). Testing
+requires mocking the entire execution environment (GitHub Actions context, CLI
+installation, process lifecycle).
+
+**Risk**: Very low. The Forgejo token path calls `setupForgejoToken()` which
+has 8 dedicated tests.
+
+**Coverage path**: Integration tests in the separate `install-test` repo.
+
+## Known Issue: CLAUDE_APP_BOT_ID Discrepancy
+
+`src/github/operations/comments/create-initial.ts` (line 23) defines a local
+`CLAUDE_APP_BOT_ID = 209825114`, while `src/github/constants.ts` exports
+`CLAUDE_APP_BOT_ID = 41898282`. These are different values with the same name.
+This should be investigated in M5 (code quality review).