AI Agent Kit / Control plane
Open-source engineering platform
Give AI agents room to work. Keep control.
Repository context, explicit approval, and evidence in one workflow.
Repo context
Plan
Policy gate
Evidence
AllowAskDeny
Evidence readynpx --yes @hunpeolabs/ai-agent-kit@latest bootstrap01 / Why
Fast agents need firm boundaries.
Speed without repository context, permission, and proof becomes risk.
- 01UnderstandRead the repository.
- 02PlanDefine the change.
- 03AuthorizeAsk before risk.
- 04ExecuteStay in scope.
- 05VerifyRecord evidence.
02 / What it controls
Context. Permission. Proof.
- Read the repoUse repository intelligence and local constraints.
- Plan the changeMake expected edits and checks reviewable.
- Ask before riskApply Allow, Ask, or Deny decisions.
- Act within scopeTie execution to the approved diff.
- Leave evidenceKeep receipts another reviewer can inspect.
Evidence
Every action leaves a receipt.
Authorization, verification, and the resulting change stay connected.
Evidence (JSON)
{
"policy": "ask",
"status": "tested",
"evidence": "recorded"
}Approval (diff)
+ validate(input)
+ audit.log("approved")
+ tests.verify()Test results
PASS policy.test PASS validation.test Evidence ready
03 / Current boundary
Open source. Inspectable by design.
An engineering platform—not a promise of every production outcome.
Where is the source?
The verified public source is available on GitHub.
How do I install it?
Run the bootstrap command shown above from the repository you want to govern.
How do I report an issue?
Use the public GitHub issue tracker so the report stays inspectable.
What support is available?
Use the Hunpeo Labs contact page for product or implementation questions.