Reflective is a memory-native coding harness for teams using AI agents in the terminal.
Most coding agents can inspect a repository, run commands, and propose patches. Reflective is built for the next layer of work: helping agents understand the context a strong engineer would ask for before changing code.
That context includes decisions, incidents, ownership, local conventions, test recipes, deployment constraints, failed attempts, and team-specific workflows. Reflective brings that knowledge into the coding session in a way that is reviewable, permission-aware, and reusable.
Reflective is coming soon. Star this repo to follow public updates and beta availability.
AI coding gets harder inside real teams. The agent needs to know more than the current prompt and the current file tree.
Reflective is designed around five product principles:
- Context before action: compile the relevant repo, memory, decision, and test context before planning or editing.
- Memory that compounds: preserve useful lessons from previous work so the next session starts smarter.
- Explicit workflows: make repeatable team practices available as skills instead of relying on tribal knowledge.
- Controlled automation: use hooks, permissions, checkpoints, and traces so automated work stays inspectable.
- Terminal-native execution: keep the workflow close to the tools engineers already use to read code, run tests, and review diffs.
Reflective is intended to be a focused CLI experience:
reflInside a session, you describe the work in plain English. Reflective prepares context, plans against the actual repository, uses approved tools, records what happened, and reports verification evidence before calling work complete.
Example workflows:
- investigate a failing test with prior incident context;
- ask which files and owners are relevant before making a risky change;
- run a skill for release notes, migrations, or review prep;
- use hooks to enforce local policy before write tools run;
- turn a successful trace into reviewable team memory.
| Surface | Purpose |
|---|---|
refl |
terminal entrypoint |
| Memory | reviewed knowledge from past sessions, decisions, and incidents |
| Skills | reusable workflows for common engineering tasks |
| Hooks | local lifecycle checks and context injection |
| Traces | evidence of actions, tool use, diffs, and verification |
Reflective is currently invite-only. Public installation is coming soon.
Public beta details, install commands, and release notes will be published in this repository.
We are especially interested in workflows where existing coding agents lose important team context.