Drain stdin in generated hook commands to avoid BrokenPipeError

[LeonCode2025]

Summary

Fixes #493.

Generated Codex and Claude hook commands currently do not consume hook event JSON from stdin before running code-review-graph commands. When a hook event contains a large payload, the hook command can exit successfully while the caller is still writing stdin, which causes Broken pipe / EPIPE errors in Codex.

This PR makes the generated hook commands explicitly drain stdin before running the existing git and code-review-graph commands.

Changes

• Prefix generated Claude hook commands with cat >/dev/null || true;
• Prefix generated Codex hook commands with cat >/dev/null || true;
• Apply the stdin drain to both PostToolUse and SessionStart generated commands
• Add a regression test that writes a large stdin payload to the generated PostToolUse command and verifies it exits successfully without BrokenPipeError

Why

The hook command itself may still return exit code 0, but the parent process can fail while writing hook stdin if the hook exits before consuming the payload. Draining stdin keeps the hook behavior non-blocking while avoiding the caller-side broken pipe error.

Cursor/Gemini hook scripts already consume stdin, so this also makes the Codex/Claude generated hooks consistent with the existing integrations.

Testing

• Ran pytest tests/test_skills.py
• Verified the new large-stdin regression test passes
• Verified the generated hook command exits with code 0 and no BrokenPipeError