Add git-ai blame visualization to JetBrains plugin

Claude.md

@@ -1,5 +1,67 @@
 - In plan mode, always use the /ask skill so you can read the code and the original prompts that generated it. Intent will help you write a better plan
 
+## Pre-Commit Checks (MANDATORY)
+
+Run all applicable checks **before every commit**. Do not commit if any check fails.
+
+### Rust (`src/`)
+- Run `cargo fmt -- --check` — CI enforces this with `-D warnings`. Run `cargo fmt` to auto-fix.
+
+### IntelliJ plugin (`agent-support/intellij/`)
+
+**Run the lint script** on the changed directories:
+```bash
+bash ~/.claude/projects/-Users-kcbalusu-Desktop-Project-git-ai/scripts/intellij-lint.sh <changed-directory>
+```
+
+**Manually verify these patterns** (not all are caught by the lint script):
+
+#### Process I/O — Pipe deadlock prevention
+When spawning external processes that use stdin/stdout/stderr:
+1. **Never write stdin synchronously before starting stdout/stderr readers.** If content exceeds the OS pipe buffer (~64KB), the parent blocks on stdin write while the child blocks on stdout write — deadlock. Start all stream readers (via `CompletableFuture.supplyAsync`) **before** writing to stdin, and write stdin asynchronously too.
+2. **Never read stdout/stderr after `waitFor()`.** Same deadlock from the other direction.
+3. Correct flow: `start process → start stdout reader → start stderr reader → write stdin async → waitFor → collect results`.
+
+#### IntelliJ Platform API — Common misuses
+1. **`fileOpened` callback: never use `selectedTextEditor`** — it returns the focused editor, not the editor for the `file` parameter. Use `source.getEditors(file).filterIsInstance<TextEditor>()` instead.
+2. **Shared state across editors:** debounce timers, alarms, or schedulers must be **per-editor**, not shared. A shared `cancelAllRequests()` or `cancel()` cancels pending work for all editors.
+3. **`Disposer.dispose()` for editor state** must be called inside `invokeLater` to avoid threading races with EDT.
+4. Verify correct API signatures — missing abstract method overrides, deprecated API usage, wrong method parameter types.
+
+### All code
+- Check for compilation errors, API compatibility issues, thread safety, concurrency bugs
+- Review for race conditions and shared state issues
+
+## PR Workflow (MANDATORY)
+
+### 1. Always create PRs in draft mode
+Use `gh pr create --draft` for every PR. Never create a non-draft PR.
+
+### 2. Thorough code review before marking ready
+After creating the draft PR, spin off a sub-agent to do an **extremely thorough review** of the entire changeset:
+- Run all pre-commit checks above
+- Fix any issues found, amend the commit, and push
+
+**Loop this review-fix cycle until zero issues are found.**
+
+### 3. Monitor CI after push (every 5 minutes)
+After pushing to the draft PR, poll CI status every 5 minutes using:
+```bash
+GH_CONFIG_DIR=/tmp/gh-cfg gh run list --repo git-ai-project/git-ai --branch <branch> --limit 1 --json status,conclusion
+```
+When a CI run completes:
+- If **failed**: fetch the logs, identify the error, fix it, push, and continue monitoring
+  ```bash
+  GH_CONFIG_DIR=/tmp/gh-cfg gh run view <run-id> --repo git-ai-project/git-ai --log-failed
+  ```
+- If **passed**: report success to the user. Do NOT mark the PR as ready — the user will do that.
+- Distinguish between **our failures** (compilation, verification) and **pre-existing/unrelated failures** (E2E opencode fish_add_path, benchmark margin noise)
+
+**ALL review, fixing, and CI monitoring MUST happen while the PR is in draft mode.**
+
+### 4. GH CLI access
+Use `GH_CONFIG_DIR=/tmp/gh-cfg` prefix for all `gh` commands to avoid sandbox config permission issues. The `GH_TOKEN` environment variable provides authentication.
+
 ## Task Master AI Instructions
 **Import Task Master's development workflow commands and guidelines, treat as if import is in the main CLAUDE.md file.**
 @./.taskmaster/CLAUDE.md