Workflow Launchpad

Use this to start work cleanly and avoid context drift.

Project Bootstrap

1. Run $repo-intelligence-brief if the repo is unfamiliar.
2. Run $low-token-context to create a compact task brief.
3. Create .codex-memory/decisions.md only if durable decisions are expected.
4. Choose one active workflow:
   • build
   • debug
   • review
   • research
   • experiment
   • MCP integration
5. Define verification before editing.

Control Plane Setup

Use this when the project has daemons, background workers, queues, MCP/tool servers, release tasks, API-to-tool bridges, or parallel agents.

## Control Plane
Command surface:
Runtime state:
Lifecycle owner:
Ports/namespaces:
Status/logs:
Workers/queues:
Agent scopes/leases:
Trust boundaries:
Verification:
Stop/recover:

• Write down the status/log/health command before starting processes.
• Keep runtime state, credentials, caches, and local databases out of source control.
• Assign write scopes or leases before parallel agent work.
• Run $mcp-safety-review before trusting side-effecting tool surfaces.
• Escalate production traces, evals, guardrails, budgets, shadow traffic, or release gates to $auto-skill-build-agent-reliability-loop.

Daily Rhythm

# Today
## Goal
## Context
## Active Files
## Commands
## Risks
## Done Criteria

End-Of-Session

• Summarize decisions into memory.
• Record failed attempts only if they prevent repeated work.
• Update task brief with next action.
• Keep logs out of memory unless compacted.

Use $codex-supercharge-router when the next workflow is unclear.

Read plugins/codex-supercharge/references/workflow-control-plane.md for the source-backed control-plane pattern.

Skip When

• The task already has a precise plan and validation command.
• The user only wants an immediate one-off answer.

Validation

• The active workflow, verification command, and done criteria are written down.
• Any durable decisions are placed in memory, not left only in chat.
• The chosen skill path matches the current repo/task risk.
• Daemon, worker, queue, MCP/tool, or multi-agent work has an explicit control plane before execution.