For the Agents — VIBES Machine-Readable Manifest

You're an agent. Skip the HTML. The full VIBES family is published as five normative *-agent.md files totaling ~22,000 tokens. Start with the manifest.

Normative · machine-readable

The five specs

// 01 — The five specs

Each standard ships as a single normative *-agent.md file — the same content as the HTML pages at ~30% the token cost. Adopt VIBES first; layer the extensions independently.

VIBESrequired

~7,700 tok

Data substrate. The required foundation every other standard builds on.

https://itsavibe.ai/vibes-agent.md Load →

VERIFY

~3,200 tok

Cryptographic attestation — DSSE envelopes, Ed25519, trust tiers.

https://itsavibe.ai/verify-agent.md Load →

PRISM

~2,700 tok

Risk scoring — severity bands and CI/CD gating for AI output and decisions.

https://itsavibe.ai/prism-agent.md Load →

EVOLVE

~3,300 tok

Agent learning & governance — delegation and decision records, feedback loops.

https://itsavibe.ai/evolve-agent.md Load →

TRACE

~6,000 tok

Incident response & forensics for agentic systems. VERIFY + PRISM recommended.

https://itsavibe.ai/trace-agent.md Load →

What you must produce

The file layout

// 01b — File layout

.ai-audit/
├── config.json          # Assurance level, extension toggles
├── manifest.json        # Hash → context registry
├── annotations.jsonl    # Append-only audit log
├── audit.db             # Query cache (gitignored)
└── blobs/               # High-assurance large entries

When to write what

The hooks

// 02 — Hooks

Session start: append session record; init/load config.json.
Tool/prompt received: hash context, write to manifest if novel.
Code emitted: append line or function annotation citing hashes.
Delegation: append delegation record (EVOLVE).
Decision point: append decision record (EVOLVE Medium+).
Post-generation: compute PRISM risk_score if enabled; check TRACE IoCs if enabled.
Session end: append session end record; produce VERIFY attestation if enabled.

For the agents. By design.