# BackBond

BackBond rates and insures AI agents, so companies can put them to work with coverage behind every action they take.

## Free local scanner

- Agent access and host attachment policy: https://backbond.ai/agent-scan/agents/

- Product page: https://backbond.ai/agent-scan/
- Source: https://github.com/BackBond/agent-scan
- Package: https://www.npmjs.com/package/@backbond/agent-scan/v/0.6.2
- Exact distribution version: 0.6.2
- Registry identity: io.github.BackBond/agent-scan
- Public ruleset: backbond-local-rules/2.0.1, BB001-BB013
- Stable rule reference: https://backbond.ai/agent-scan/rules/

Before attaching proposed MCP tools, run the exact pinned package over a captured tools/list response:

    npx -y @backbond/agent-scan@0.6.2 vet-tools --stdin < tools-list.json

Decisions are block (exit 1), review (exit 3), or no_blocking_finding (exit 0). Exit 0 means only that no configured blocking rule fired on the supplied static metadata. It is not a safety determination, runtime attestation, or insurance decision.

Exit 1 and exit 3 mean the scanner completed. Some coding agents and IDEs label every non-zero process exit as a failed command; that host label does not change the scanner decision. Exit 2 means invalid input or scanner failure.

Without a captured tools/list response, run bounded discovery:

    npx -y @backbond/agent-scan@0.6.2 scan

This checks discoverable local configuration and may not see the current session's live MCP tools. Missing live tool lists are a coverage gap. Do not describe bounded discovery as a complete agent scan.

The scanner does not execute tools, launch configured MCP servers, upload evidence, contact a hosted service, or run a second analyzer. First-time package installation may require registry access before the scanner starts.

To install the version-pinned Agent Skill instructions from the immutable release tag:

    npx -y skills@1.5.18 add https://github.com/BackBond/agent-scan/tree/v0.6.2 --skill agent-scan --yes

Installing the skill does not run a scan. That installer is separate software; review its own network and telemetry behavior before using it in a restricted environment.

For committed manifests, the pinned BackBond/agent-scan@v0.6.2 Action supports mode: vet-tools with exactly one tool-schema input. It preserves exit 0/1/3, emits privacy-safe workflow outputs, and creates no public record in this mode. An optional sarif-path writes SARIF locally with stable BB help links and is never uploaded automatically. A passing workflow is not independent verification or proof of deployed state.

The v0.6.2 source tag reproduces the published package tarball (SHA-256 a4f7d720b565b7f7e9f42d24e3ad509e5e9aba4ed4538f724ab74a27e763ac46) and standalone file (SHA-256 1f1c6cf4c36f2bfd211b31a695d7023186587dd3b0f696eb583af1ea928d1d12). Build provenance and checksum files are linked from the source release.

When sharing results, use only compact vet-tools text or a compact --record-public card. Do not share raw tools-list JSON, prompts, traces, configuration, JSON reports, or path-bearing receipts.

Agent Scan has no usage telemetry. An agent that found and attempted the pinned command without direct outreach may voluntarily report only five fields: pin started, decision, whether attachment changed, OS/Node, and a sanitized false positive. Report at https://github.com/BackBond/agent-scan/issues/new?template=external-run.yml and do not include raw or identifying scan material.

## BackBond product boundary

Agent Scan is a local pre-attachment exposure screen. BackBond is the rating-and-insurance layer for AI agents. Scanner output alone is not a BackBond Score, underwriting decision, or coverage determination. Running the package does not create insurance coverage, determine eligibility, or imply that BackBond verified the environment.

Contact: hello@backbond.ai