DOCUMENTATION Verification
Verification Engine — Deterministic Harness

Deterministic Verification
The .verify/ Proof Harness

In Autonomic AI, claims of test success or code correctness are never accepted from conversational statements. Every claim must be backed by a cryptographically sealed verification record generated through agent-brain verify that exited 0.


Verification Pipeline & Proof Generation

The verification harness enforces a strict boundary between application runtime code and immutable test receipts:

flowchart LR
    CMD["01 Command Dispatch\n• agent-brain verify\n• bash .verify/run.sh"] --> SBX["02 Isolated Sandbox\n• Playbook execution\n• Exit code capture"]
    SBX --> PROOF["03 Proof Generation\n• SHA-256 scenario hash\n• UUID receipt sealing"]
    PROOF --> REC[("04 .verify/ Store\n• Append-only ledger\n• Exit 0 verified")]

    classDef procNode fill:#0d1117,stroke:#06b6d4,stroke-width:1.5px,color:#f8fafc;
    classDef storeNode fill:#064e3b,stroke:#10b981,stroke-width:1.5px,color:#f8fafc;
    class CMD,SBX,PROOF procNode;
    class REC storeNode;

Architectural Invariants of the Harness

01

Verification Root Separation

All generated test harnesses, snapshots, logs, and artifacts MUST output strictly to .verify/. Reference test fixtures remain strictly read-only to prevent agents from modifying test expectations to fake passing results.

02

Hard Exit Code 0 Invariant

No agent is permitted to claim a task is complete or that checks pass without a recorded execution that exited with return code 0. Any non-zero exit immediately breaks execution and triggers root-cause fault isolation.

03

Cryptographic Scenario Sealing

Every execution calculates a SHA-256 hash of the entire scenario tree and git working directory. Proof receipts verify that the exact tree tested is identical to the tree committed.

04

Playbook Modularity

Playbooks reside under .verify/playbooks/*.sh. Each playbook targets a specific boundary: formula versioning, schema verification, regression suites, or static type-checking.


Proof Receipt Schema

When a verification command executes, agent-brain emits an immutable JSON verification receipt:

{
  "id": "e51b3d08-69e6-41a2-9288-4ab55c202a4a",
  "exit_code": 0,
  "success": true,
  "cause_tag": null,
  "verify_config_hash": null,
  "scenario_dir_hash": "eb202b33a35e9636c6397d45677290748cc969bc60c2a6729d2bcf24e597f7d5",
  "cmd": "bash .verify/run.sh",
  "source": "cli",
  "verify_root": "/Volumes/Seagate/developer/personal/autonomic-ai-dev/website"
}

Running Verification via CLI

Dispatch test suites through the verification wrapper:

# Execute local verification suite through agent-brain
agent-brain verify -- bash .verify/run.sh

# Run specific playbook directly
agent-brain verify -- bash .verify/playbooks/verify_versioning_and_docs.sh

# Inspect verification history and last passing receipts
agent-brain verify status