Context Management
Scannednpx machina-cli add skill a5c-ai/babysitter/context-management --openclawContext Management
Project context loading, isolation, and persistent state management across CCPM sessions.
Capabilities
- Load project context from .claude/ directory structure
- Maintain persistent context across sessions (PRD, epic, task states)
- Isolate agent contexts per work stream
- Context handoff between phases
- Session state recovery
Context Sources
.claude/
prds/ # PRD documents
epics/ # Epic and task documents
agents/ # Agent definitions
context/ # Project-wide context
commands/ # Command definitions
rules/ # Reference rules
Context Flow
- Phase 1 (PRD): Loads project description, creates PRD context
- Phase 2 (Epic): Loads PRD, creates architecture context
- Phase 3 (Tasks): Loads PRD + Epic, creates task context
- Phase 4 (Sync): Loads all artifacts, syncs to GitHub
- Phase 5 (Exec): Loads stream-specific context per agent
Isolation
Each parallel agent receives only the context relevant to its stream, preventing context pollution and keeping agents focused.
Source
git clone https://github.com/a5c-ai/babysitter/blob/main/plugins/babysitter/skills/babysit/process/methodologies/ccpm/skills/context-management/SKILL.mdView on GitHub Overview
Context Management loads project context from the .claude directory and preserves state across CCPM sessions, including PRD, epics, and task data. It isolates agent contexts per work stream to prevent cross-stream contamination and supports handoff between phases and session recovery.
How This Skill Works
The system reads from the .claude/ directory (prds, epics, agents, context, commands, rules) to assemble phase-specific contexts. It creates PRD context in Phase 1, architecture context in Phase 2, and task context in Phase 3, then syncs artifacts in Phase 4 and loads stream-specific contexts in Phase 5. It keeps contexts persistent across sessions and isolates each agent to its work stream, enabling context handoff between phases and session recovery.
When to Use It
- Starting a CCPM project and loading the initial PRD context from .claude
- Moving from Phase 1 PRD to Phase 2 Epic to build architecture context
- Running parallel agents on separate work streams and needing isolated contexts
- Recovering a session after a restart to restore PRD, epic, and task states
- Syncing all artifacts to GitHub after Phase 4 for a durable handoff
Quick Start
- Step 1: Ensure your .claude directory contains prds/, epics/, agents/, context/, commands/, and rules/.
- Step 2: Run Phase 1 to load the PRD and create the PRD context, then advance to Phase 2 to build architecture context.
- Step 3: Progress through Phases 3 and 4, verify per-stream context isolation, perform session recovery if needed, and sync artifacts to GitHub.
Best Practices
- Use the .claude directory structure consistently (prds, epics, agents, context, commands, rules)
- Ensure each work stream receives only its own context to maintain isolation
- Validate handoffs between phases to preserve context continuity
- Enable and test session state recovery to handle interruptions gracefully
- Keep PRD, Epic, and Task documents up to date to reflect the current context
Example Use Cases
- CCPM implementation for a mobile app with separate streams for design, development, and QA
- Multi-agent orchestration across epics and tasks with isolated contexts
- Phase 2 architecture context creation after Phase 1 PRD load
- GitHub artifact syncing after Phase 4 for traceable handoffs
- Session recovery after a restart to restore PRD, epic, and task states