gala-team-e2elisted

# gala_team end-to-end test driver Run real-claude end-to-end scenarios against the gala_team binary, then produce a comprehensive analysis covering UX, user flows, code quality, bug detection, JSON leaks, and usability. Used to catch regressions that synthetic test-driver replays miss — gala's JSON codec, claude's stream-json output, and the orchestrator FSM all have edge cases that only show up with real subprocesses. **Argument:** `$ARGUMENTS` — Optional comma-separated scenario list (default: run all). Available scenarios: - `solo` — Single lead, real claude edits + commits, gh pr create. - `2eng_qa` — 2 engineers + QA, parallel dispatch, deliverable manifest, summary. - `consult` — Cross-team `@consult` between two repos. - `recovery` — Start fresh / Esc / quit-reopen leftover-data check. - `worktrees` — `worktree-per-engineer` mode + cleanup on Start fresh. - `leak` — Synthetic scenarios that historically tripped JSON-leak guards. ## Prerequisites Run from the gala_team repo root. The skill needs: - `bazel` on PATH and a built `gala_team` + `gala_team_headless` binary - `claude` on PATH and authenticated - `node` on PATH (used to extract JSON fields from headless output) - `git` on PATH Build first if either binary is stale: ```bash bazel build cmd:gala_team cmd/gala_team_headless:gala_team_headless ``` ## Instructions ### Step 1 — Sanity-check the build Run `bazel test app/...`. If anything fails, **stop** and report — broken unit tests mean the binary can't be