Skip to content
Back to skills

2486 Dogfood Bugs 606deb02

ASecurity

Tracking bugs discovered during production dogfooding runs.

  • 9 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 11, 2026
devopspythongodockergit

Works with

  • claude code

Security analysis

A100/100

Scanned October 11, 2026

npx -y skills add tools-only/X-Skills --skill 2486-dogfood-bugs_606deb02 --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of 2486 Dogfood Bugs 606deb02?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for 2486 Dogfood Bugs 606deb02
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/tools-only-2486-dogfood-bugs-606deb02/badge)](https://www.skillsdirectory.com/skills/tools-only-2486-dogfood-bugs-606deb02)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

SKILL.md
# ZERG Dogfood Bug Tracker

Tracking bugs discovered during production dogfooding runs.

## Categories

| Category | Description |
|----------|-------------|
| orchestrator | Task scheduling, level management, main loop |
| worker | Task execution, Claude invocation, context tracking |
| merge | Branch merging, conflict resolution, rebase |
| state-ipc | State file read/write, worker-orchestrator sync |
| launcher | Process/container spawning, health checks |
| plugin | Plugin loading, hooks, gates, lifecycle events |

## Severity Levels

| Level | Description | Response |
|-------|-------------|----------|
| P0 | Blocker — stops rush execution | Fix immediately |
| P1 | Major — task failures, data loss | Fix before next rush |
| P2 | Minor — cosmetic, non-blocking | Fix when convenient |

## Bug Template

### BUG-XXX: [Title]
- **Severity**: P0/P1/P2
- **Category**: orchestrator/worker/merge/state-ipc/launcher/plugin
- **Discovered**: YYYY-MM-DD
- **Status**: Open/In Progress/Fixed
- **Feature**: [feature name]
- **Repro**: [steps to reproduce]
- **Expected**: [expected behavior]
- **Actual**: [actual behavior]
- **Root Cause**: [analysis]
- **Fix**: [commit hash or PR]

---

## Bugs

### BUG-001: /zerg:rush --container bypasses Orchestrator pipeline
- **Severity**: P0
- **Category**: orchestrator
- **Discovered**: 2026-01-29
- **Status**: Open
- **Feature**: production-dogfooding
- **Repro**: Run `/zerg:rush --workers 5 --container` in a Claude Code session
- **Expected**: The rush command invokes `Orchestrator.start()` with `launcher_mode="container"`, which spawns Docker containers via `ContainerLauncher`, creates git worktrees, runs workers through the full `WorkerProtocol` pipeline, and coordinates level-by-level execution with merge gates.
- **Actual**: The rush command dispatches tasks as Claude Code Task sub-agents, completely bypassing the Orchestrator, ContainerLauncher, git worktrees, state IPC, and merge coordination. No Docker containers are started regardless of the `--container` flag.
- **Root Cause**: The `/zerg:rush` command documents three execution modes (lines 349-413 of `zerg:rush.md`): container, subprocess, and task-tool. Auto-detection logic (lines 356-359) selects task-tool mode when running as a Claude Code slash command (rule 3), even when `--container` is explicitly passed. Rule 1 ("If `--mode` is explicitly set → use that mode") should take precedence, but the executing agent ignored the explicit flag and defaulted to task-tool mode.
- **Fix**: Two changes needed: (1) The executing agent must honor explicit `--mode`/`--container` flags per rule 1 of the auto-detection logic. When `--container` is specified, invoke `Orchestrator(feature=feature, launcher_mode="container").start(task_graph_path, worker_count)` via Python subprocess instead of dispatching Task sub-agents. (2) Add a guard in the rush command file that explicitly rejects task-tool mode when `--container` or `--mode container` is passed.

### BUG-002: No verification that container mode was actually used
- **Severity**: P1
- **Category**: launcher
- **Discovered**: 2026-01-29
- **Status**: Open
- **Feature**: production-dogfooding
- **Repro**: Complete a rush with `--container` flag
- **Expected**: Post-rush status or logs should confirm container mode was active (container IDs, Docker network, worktree paths)
- **Actual**: No diagnostic output confirms or denies container mode usage. The user had to explicitly ask `/zerg:debug` to discover containers were never used.
- **Root Cause**: The rush completion output does not include launcher mode verification. Neither `/zerg:status` nor the rush summary reports which launcher mode was active.
- **Fix**: Add launcher mode to rush summary output and `/zerg:status` response. Include container IDs in worker status when container mode is active.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…