Agent Action Audit Template

A clean-room public template for documenting and testing agent action audit records.

This repository uses synthetic examples only. It does not include private prompts, production logs, customer data, proprietary architecture, real receipts, internal governance labels, credentials, or deployment details.

Release: v1.0.0

Purpose

Agent systems can be useful only if their actions can be reviewed, bounded, and reconstructed. This template shows one small public pattern for representing safe and blocked agent actions with fake receipts and human review metadata.

What This Demonstrates

• audit trail design
• safe versus blocked action records
• human review checkpoints
• JSON Schema-backed receipt validation
• negative tests for invalid or incomplete receipts
• public-safe synthetic examples
• pytest verification

Included

• one synthetic allowed-action receipt
• one synthetic blocked-action receipt
• action receipt JSON schema
• human review JSON schema
• pytest validation
• GitHub Actions test workflow
• synthetic examples only

Validation Coverage

The test suite checks that:

• valid allowed-action receipts pass schema validation
• valid blocked-action receipts pass schema validation
• missing required fields fail validation
• invalid decision enum values fail validation
• unexpected top-level fields fail validation
• missing nested human-review fields fail validation
• invalid nested human-review status values fail validation
• examples avoid prohibited private-data markers

Repository Structure

agent-action-audit-template/
├── .github/
│   └── workflows/
│       └── test.yml
├── docs/
│   ├── audit_log_design.md
│   └── human_review_flow.md
├── examples/
│   ├── blocked_action_receipt.json
│   └── safe_action_receipt.json
├── schemas/
│   ├── action_receipt.schema.json
│   └── human_review.schema.json
├── tests/
│   └── test_receipt_schema.py
├── README.md
└── requirements.txt

Run Locally

pip install -r requirements.txt
pytest

Limitations

This is a small template, not a production governance system. Passing these tests does not prove agent safety, deployment readiness, compliance, or complete auditability.

Public Boundary

This repository intentionally excludes:

• private prompts
• private schemas
• production logs
• customer data
• real receipts
• proprietary architecture
• internal governance labels
• deployment details