[■]─[▪] sidecar
Get Started
[■]─[▪]
Codex Workflow

Codex can execute the task. Sidecar preserves the project memory.

Use Sidecar with Codex to capture what happened and why, directly inside your repo, so the next human or agent can continue without re-learning the same context.

terminal bash 
sidecar context --format markdown
# ... implement task with Codex ...
sidecar summary refresh

Why Codex workflows need a memory layer

In Codex-heavy workflows, tasks are often completed in discrete sessions. Without explicit project memory, each new session repeats discovery work and may miss prior assumptions.

What to record after each Codex task

Decision

Record the tradeoff if behavior, architecture, or design changed.

Work log

Capture what was finished and the files that changed.

Task

Add unresolved follow-ups instead of leaving them in chat only.

Summary

Refresh global project context for the next kickoff.

Codex-friendly end-of-task checklist

  1. Run sidecar context --format markdown before starting.
  2. Implement and verify the code change.
  3. Run sidecar worklog record with outcomes and touched files.
  4. Add sidecar decision record when significant choices were made.
  5. Add sidecar task add for follow-up items.
  6. Run sidecar summary refresh.

Example Codex handoff

handoff snapshot bash 
sidecar worklog record --done "Added retry middleware and circuit-breaker fallback" --files src/http/retry.ts,src/http/client.ts --by agent
sidecar decision record --title "Prefer middleware retries over call-site retries" --summary "Central policy gives consistent behavior and less duplicated logic" --by agent
sidecar task add "Add retry metrics panel in local UI" --priority medium --by agent
sidecar summary refresh

Where this fits in the stack

Sidecar is a companion to Codex, AGENTS.md, and Git. It does not replace source control or prompt discipline. It gives Codex sessions a stable, repo-local place for context that should outlive one run. See command reference for full options.

FAQ

Does Sidecar make Codex remember everything automatically?

No. Sidecar provides structured context that Codex can read and update. Memory quality depends on using the record commands consistently.

Should I record a decision after every small code edit?

Not necessarily. Record decisions when behavior, architecture, or design tradeoffs changed in a way future contributors should understand.

Can humans use the same records created by Codex?

Yes. Sidecar records are intended for both humans and agents in the same repository workflow.

Keep Codex sessions connected through shared repo memory.

Git tracks code changes. Sidecar tracks rationale, decisions, and follow-up context for future Codex sessions and human reviewers.

Get Started Agent Workflow Guide