Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
No functional or protocol changes detected in this release. - Updated skill description to emphasize Codifica as shared, persistent memory for agent/human collaboration. - No modifications to protocol instructions or skill logic.
What This Skill Does
Codifica is a file-based protocol for coordinating tasks between AI agents and humans in a shared Git repository. Task state, ownership, decisions, and handoffs are stored as plain text files committed alongside the code. Git serves as the audit log with no external service required.
Because state lives in Git, every task transition is versioned, diffable, and auditable without standing up a separate coordination service.
When to use it
- Handing off an in-progress task from one agent to another without losing context
- Tracking which agent claimed a task and when across a multi-agent pipeline
- Reviewing what work was completed on a feature without reading chat logs
- Preventing two agents from claiming the same task simultaneously
- Enforcing task dependencies so agents don't start work before blockers are resolved
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: implement the authentication feature in this repo
- 1Read codifica.json at the repo root to load protocol config and locate the state file
- 2Read the spec file referenced in codifica.json for project rules and task definitions
- 3Scan work.md for tasks with state todo and matching owner or unassigned, respecting depends_on
- 4Claim the matching task in a single atomic commit: set state to in_progress, record owner and claimed_at, push immediately
- 5Complete the work per the task's acceptance criteria, then update the task with execution_notes, artifacts, and move state to to_be_tested
Task committed as in_progress with claim metadata, work completed, and task moved to to_be_tested with a summary execution note and artifact paths recorded