feat(agents): add --status flag to display live session info in Claude Code status bar

[vadimvlasenko]

Summary

Adds a new --status flag to codemie-claude that enables a live status bar inside Claude Code, showing real-time session info: AI model, current directory, git branch, context window usage, cost, and elapsed time.

Changes

• New --status CLI flag — added to AgentCLI and wired into the Claude plugin lifecycle
• Statusline script (codemie-statusline.mjs) — standalone Node.js script deployed to ~/.claude/ at session start; reads Claude Code's JSON hook payload via stdin and renders a two-line status bar with ANSI colors
• Session lifecycle integration — beforeRun deploys the script and injects statusLine config into ~/.claude/settings.json; afterRun restores the original settings after the session ends
• Supported version bump — CLAUDE_SUPPORTED_VERSION updated from 2.1.31 to 2.1.41
• Unit tests — 13 tests covering beforeRun and afterRun hooks (script deployment, path quoting, settings injection/cleanup, parse error safety, flag idempotency)

Impact

# Enable live status bar for a session
codemie-claude --status

# Status bar shows two lines inside Claude Code:
# [Claude Sonnet 4] 📁 my-project | 🌿 feature/my-branch
# ████████░░ 82% | $0.0142 | ⏱️ 3m 12s

• No impact on sessions that don't pass --status
• Settings are restored to their original state after each session

Checklist

• Self-reviewed
• Manual testing performed
• Documentation updated (if needed)
• No breaking changes (or clearly documented)

[8nevil8]

Code Review — Approved

Files reviewed: src/agents/core/AgentCLI.ts, src/agents/plugins/claude/claude.plugin.ts, src/agents/plugins/claude/plugin/codemie-statusline.mjs, src/agents/plugins/claude/__tests__/claude.plugin.statusline.test.ts, scripts/copy-plugins.js

---

Summary

The --status flag implementation is clean, well-scoped, and follows project conventions correctly. The code introduces a live status bar for Claude Code sessions with proper lifecycle management, cross-platform guards, and good error handling. All critical areas pass review.

---

What was checked

Cross-Platform Compatibility — The chmod call is correctly gated behind process.platform !== 'win32', and path.join() is used consistently throughout. The statusline script itself uses execSync with a stdio: ['pipe', 'pipe', 'ignore'] option that works cross-platform. No hardcoded path separators.

Architecture Adherence — The feature lives entirely within the Claude plugin's lifecycle hooks (beforeRun/afterRun), which is the correct layer for session-scoped setup. No layer violations.

Settings Safety — The beforeRun hook correctly parses settings.json before writing and bails out with a warning on malformed JSON, preventing data loss. The afterRun hook restores original state and handles the no-settings case gracefully.

Module-level flag design — Using statuslineManagedThisSession at module scope (instead of an env var) is the right call — it keeps internal state out of subprocess environments. The flag is reset to false at the start of afterRun, so a second call is correctly a no-op.

Security — sanitizeLogArgs is used on all logger.warn calls that include path or error context. No credentials or sensitive data are logged.

Error Handling — Uses logger.warn (not logger.error or throw) for non-critical failures like parse errors and cleanup failures, which is appropriate since the status bar is an optional feature. Silent fail in the statusline script itself is correct — crashes must not propagate to Claude Code.

Standalone script justification — The comment in codemie-statusline.mjs explaining why child_process and path are imported directly (the script runs in isolation without the project runtime) is accurate and appreciated.

Previously raised comment — The redundant copyFiles loop flagged by 8nevil8 on scripts/copy-plugins.js has been fully addressed: codemie-statusline.mjs is now in the plugin/ subfolder and is picked up by the existing cpSync, with the copyFiles loop and copyFileSync import removed entirely. Resolved the thread.

Tests — 13 tests cover the meaningful paths: no-op when flag is absent, script deployment, directory creation, path quoting, idempotency on existing statusLine, malformed JSON safety, afterRun cleanup, double-call no-op, and cleanup failure logging. Test isolation via vi.resetModules() in beforeEach is correct for resetting the module-level flag.

---

Positive observations

• The module-level flag approach for session tracking is a thoughtful alternative to polluting subprocess env vars.
• Quoting the script path in the command field (node "${scriptPath}") correctly handles home directories with spaces.
• The afterRun cleanup is unconditional on exit code, ensuring settings are always restored even after failed sessions.
• Build integration via the existing cpSync mechanism (scripts/copy-plugins.js) is clean and minimal.

LGTM. Ready to merge.