← Back to Daily Builds

Daily build note · June 8, 2026

AgentOps Evidence Pack

A local deterministic packaging layer that combines AI tool inventory, agent toolchain safety scans, and agent-run receipts into one client-ready AgentOps evidence packet: executive...

AI Tools Build note published Public demo coming soon

What shipped

Built AgentOps Evidence Pack, a local deterministic Python CLI that turns seeded or user-supplied AgentOps evidence into a client-ready packet.

Implemented commands:

  • python -m agentops_evidence_pack sample --path /tmp/agentops-evidence-pack-demo --force
  • python -m agentops_evidence_pack build --path /tmp/agentops-evidence-pack-demo
  • python -m agentops_evidence_pack validate --path /tmp/agentops-evidence-pack-demo
  • bash verify.sh

Generated outputs:

  • agentops/evidence-pack.md
  • agentops/control-map.md
  • agentops/weekly-review-agenda.md
  • agentops/open-risks.md
  • agentops/client-one-pager.md
  • agentops/proposal-and-sow.md
  • agentops/evidence-pack.json

Architecture

  • Standard-library Python only, matching the parent Agent Receipts Desk style.
  • Self-contained parsers for Agent Ledger-style, SkillSafe Registry-style, and Agent Receipts-style JSON exports.
  • Deterministic rule-based analysis instead of model-generated prose.
  • Stable Markdown renderers and a machine-readable JSON packet.
  • Local-only workflow with no network access, provider calls, OAuth, browser automation, GitHub API, or live MCP introspection.

Trimmed scope

Intentionally not included in this MVP:

  • Hosted UI or dashboard.
  • PDF export.
  • Live SaaS connectors.
  • OAuth or API-key based ingestion.
  • Live MCP server introspection.
  • Legal, compliance, or security attestation claims.
  • Full schema compatibility across every possible prior incubator export.

Limitations

  • Input normalization supports seeded fixtures and obvious prior-build field names, not a formal versioned schema.
  • Risk scoring is deterministic and conservative, intended for operating review rather than automated enforcement.
  • Markdown output is designed for copy/paste and review, not polished visual design.
  • Pricing language is a draft sales anchor and should be adjusted per client.

Verification

bash verify.sh compiles the package and runs scripts/smoke.py, which creates a clean /tmp/agentops-evidence-pack-demo workspace, builds the packet, validates it, and checks required business signals.

Suggested next steps

  • Add fixture tests for real Agent Ledger and SkillSafe exports once available.
  • Add optional config for company name, pricing anchors, and default owners.
  • Add a packet diff command for monthly reviews.
  • Add optional PDF rendering after Markdown content stabilizes.
  • Add intake templates for clients who do not yet have all three evidence sources.