Multipowers is a tool project that helps vibe-coding agents (Claude Code and similar environments) deliver changes in a predictable way: context first, then workflow, then role execution.

• This repo (conductor/) defines how to build and evolve the Multipowers tool itself.
• Template repo (templates/conductor/) scaffolds context for building the user's target app.

1. conductor (setup + new track): project-level stable context and track lifecycle.
2. superpowers (methodology): explicit software workflow for major changes.
3. oh-my-opencode (role selection): task-aware role dispatch for execution nodes.
4. Claudecode-Codex-Gemini (multi-CLI bridge): non-interactive claude / codex / gemini invocation per role.

Router is the coordinator. It routes by task type:

• Fast Lane: skip skills; directly dispatch an execution role for small, low-risk, bounded tasks.
• Standard Lane: choose a workflow, not a role. Example workflows: brainstorming, writing-plans, subagent-driven-development, executing-plans.

In Standard Lane, each workflow has a default executor role, but specific nodes can require specialist roles. Example: in subagent-driven-development, implementation may default to Coder, while code review nodes are forced to Architect.

• router → local/system coordinator in main Claude Code session
• architect → gemini CLI (planning/architecture + review/verification)
• coder → codex CLI (implementation)
• librarian → gemini CLI (fast research)

All external calls go through bin/ask-role, which injects context and forwards role-specific prompts to connectors.

For significant modifications, use this checklist:

1. Record changed files (e.g., git diff --name-only).
2. Run post-change checks:
   • semgrep for security/pattern issues
   • biome for TS/JS formatting/linting
   • ruff for Python linting/format checks
3. Fix findings and re-run until clean (or explicitly justify exceptions).
4. Update documentation based on changed-file scope (design/workflow/usage docs).

npm install
./bin/multipowers init --repair
./bin/multipowers doctor
./bin/multipowers track new my-feature

./bin/multipowers track new <feature-name>
./bin/multipowers track start <track-name>
# execute via fast lane or standard lane
./bin/multipowers track complete <track-name>

# choose lane only (decision)
./bin/multipowers route --task "Refactor auth boundary" --risk-hint high --json

# route + execute in one command
./bin/multipowers run --task "Refactor auth boundary" --risk-hint high --json --allow-untracked

# run standard-lane workflow (node-level role switching is defined in workflow config)
./bin/multipowers workflow run subagent-driven-development --task "Implement feature X" --allow-untracked

# inspect and validate available workflows
./bin/multipowers workflow list
./bin/multipowers workflow validate

# role dispatch bridge
./bin/ask-role architect "Brainstorm architecture"
./bin/ask-role coder "Implement task with TDD"
./bin/ask-role architect "Review for blocking issues"

# health and governance baseline
./bin/multipowers doctor
bash scripts/run_governance_checks.sh --mode strict --changed-file README.md
bash scripts/run_governance_checks.sh --mode advisory --changed-file bin/multipowers --changed-file README.md
npm run governance -- --help

# safe update workflow
./bin/multipowers update --check --json
./bin/multipowers update --apply --yes

npm test --silent

• Structured logs are written to outputs/runs/YYYY-MM-DD.jsonl.
• Router/workflow events include: lane_selected, fast_lane_dispatched, fast_lane_finished, workflow_started, workflow_node_executed, workflow_finished, governance_started, governance_finished.
• Use request_id (and optional track_id) to reconstruct one execution timeline.

multipowers/
├── bin/                     # multipowers + ask-role entry points
├── connectors/              # codex/gemini/claude wrappers
├── config/                  # default role and schema config
├── conductor/               # maintainer-facing live context/tracks
├── templates/conductor/     # scaffold for user project context/tracks
├── skills/                  # methodology skills
├── scripts/                 # validators/governance checks
└── tests/                   # regression and integration tests