Prove the Guardrails Worked: Turning a LangChain Agent's Run Into a Replay Bundle

Description and use

Cognous enables quick proof of a LangChain agent’s actions by generating a validated, signed replay bundle suitable for audit.

Problem

A CISO needs to know what an AI agent actually did during a run, not just what it was supposed to do, and must get that answer quickly.

Outcome

The agent’s run was captured, mapped to a public replay schema, validated, redacted, and signed, producing a portable evidence bundle that satisfies auditor requirements.

Workflow

• 1. Define an action manifest declaring allowed tools and review modes.
• 2. Wrap LangChain’s create_agent loop with manifest_guard middleware and RunRecorder.
• 3. Execute the agent; RunRecorder captures tool calls, decisions, and outcomes.
• 4. Export the recorder state to a ReplayBundle object.
• 5. Map Control Plane export fields to arb’s AgentReplayBundle schema.
• 6. Validate the bundle with arb validate.
• 7. Redact sensitive fields (payloads, targets, reasons, final output) using arb redact.
• 8. Sign the bundle to produce a handoff-ready evidence package.

Evidence

Cognous's Open Control Stack exists to make that question answerable in minutes instead of days.

Source

Calling recorder.generate_replay_bundle() produces a ReplayBundle object.

Source

arb validate pipeline_agent_bundle.json VALID bundle_id=138eb7ca-3da9-4de8-b197-57062d8fcf00 issues=1 WARNING W012: signature_metadata is missing.

Source

arb redact pipeline_agent_bundle.json --out pipeline_agent_bundle.redacted.json --targets --reasons --final-output Redacted bundle written to pipeline_agent_bundle.redacted.json

Source