Co-actions

A CoAction is a handoff document that captures knowledge you can’t infer from code alone. It transfers implicit knowledge between sessions and between agents. The key: it’s not a session-log dump — include only information with lasting value for the next reader.

Using it with AI agents

Ask your AI agent something like:

“Save what you’ve done so far as a CoAction”
“Create a handoff document so the next session can pick up where we left off”

The next session or another agent can then resume without needing extra context.

The detail view shows the takeaways, the content (Key Discoveries, Design Decisions, and more), and the linked plans and completion reports.

When to create one

After completing a multi-session or multi-agent project
When handing off work to another agent or team member
When implicit knowledge would be lost without documentation
Before a long pause in active development on a feature

What it contains

Section	Content
Key Discoveries	Findings not visible from code — workarounds, undocumented behavior, environment quirks
Design Decisions	The decision + rationale (why this approach, what alternatives were rejected)
Usage Scenarios	Intended use and non-obvious interaction patterns
Non-Goals	What was deliberately excluded and why (prevents scope creep)
Risks / Trade-offs	Known compromises, potential impact, technical debt
Follow-up / Known Constraints	What the next agent should continue — the most important section for handoff

Fill only the sections with meaningful content; skip the empty ones.

Source: Runner / Direct

Runner — a co-action auto-created from a runner session
Direct — a co-action created manually, without a runner

Visibility

PRIVATE (default) — only the creator can access
PROJECT — all project members can view (only the creator can change status/visibility or delete)

Takeaways

Attach key insights that emerged during the work, separate from the main content — non-obvious constraints/behaviors, undocumented decisions, workarounds to reuse (or avoid). Things the next agent might not discover on their own.

Common scenarios

Save state before ending a session — record progress and notes for the next person.
Hand off between agents — e.g. one agent finishes the backend API and hands frontend integration to another.
Checkpoint long-running work — create a co-action each session end to resume the next day.
Share a blocker — record an unresolved problem and request human review.

CoAction vs Plan

	CoAction	Plan
Purpose	Transfer work state and context	Define work intent and scope
Created by	AI agent	Human or AI
Requires approval	No	Yes (human approval)
Used	At session start/end	Before work begins