trustbundle

Pack AI agent execution traces into verifiable bundles for audit, compliance, and incident investigation.

Why

AI agents make autonomous decisions, but when something goes wrong, you need to explain what happened — not just observe or control it.

trustbundle takes agent execution traces from any source and seals them into a single, integrity-protected bundle that you can hand to auditors, attach to incident reports, or archive for compliance.

Quick Start

npm install -g trustbundle

# Initialize a project
trustbundle init

# Build a bundle from trace files
trustbundle build traces.jsonl

# Verify bundle integrity
trustbundle verify .trustbundle/bundles/<bundle_id>.json

# Show bundle summary
trustbundle show .trustbundle/bundles/<bundle_id>.json

Design Principles

1. Independence — Works standalone without agentbond or agent-trust-telemetry. Accepts any JSONL as input.
2. Single Responsibility — Build, verify, and show bundles. Nothing else.
3. Machine Verifiability — A bundle is a single self-contained file, verifiable by program. Human-readable output is the show command's job.
4. Progressive Trust — MVP uses digest (tamper detection) only. Signatures come in the next phase.
5. Adapter Design — Input sources are abstracted via adapters, so adding new sources never touches core logic.

Commands

| Command | Description | |---|---| | trustbundle init | Initialize configuration in the current directory | | trustbundle build <file...> | Build a TrustBundle from input files | | trustbundle verify <bundle.json> | Verify bundle integrity via digest re-computation | | trustbundle show <bundle.json> | Display a human-readable bundle summary |

Adapters

| Adapter | Input Format | Status | |---|---|---| | jsonl | Generic JSONL (one JSON object per line) | MVP | | agentbond | agentbond AuditRecord[] (JSON array) | MVP | | otel | OpenTelemetry spans | Planned |

Related Projects

| Project | Role | |---|---| | agentbond | Authorization & audit layer — bundle event source | | agent-trust-telemetry | Runtime contamination detection — bundle event source |

Together these form the agent trust stack: before (authorization) → during (telemetry) → after (trustbundle).

Status

v0.1.0 — MVP (digest-only integrity verification)

• Signature support planned for v0.2
• KMS integration planned for v1.0

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT