{
  "protocolVersion": "0.3.0",
  "name": "Witness-DAO",
  "description": "Governance-evidence layer for DAOs and their legal wrappers. Seals governance-significant actions — proposal, vote, multisig approval, treasury movement — to tamper-evident, externally anchored records bound to the ratified rule that authorised them, and turns that trail into evidence a regulator, an auditor or a court can read.\n\nPRODUCES EVIDENCE, NOT A COMPLIANCE CERTIFICATE. Attestation is a human act; nothing returned here constitutes legal or regulatory approval.\n\nEvery governance action carries a bindingConfidence stating how strongly it is tied to independently observable reality (onchain_confirmed > signed_offchain > api_asserted > unbound). No layer ever upgrades that grade. Verification verdicts are a closed set: ANCHORED_VALID, SIGNED_PENDING, BROKEN, INSUFFICIENT_PROOF. Never summarise SIGNED_PENDING as verified, and never collapse BROKEN (tampering) with INSUFFICIENT_PROOF (missing data).\n\nThe skills here are read-only, free and unauthenticated. Sealing is not offered remotely: it requires the caller's own Witness credential and their git-held governance rules. The liability-defence pack is deliberately not offered remotely at all, because it is a behavioural profile of an identified natural person who is usually not the requester.",
  "url": "https://witness-dao.getvda.ai/a2a",
  "preferredTransport": "JSONRPC",
  "provider": {
    "organization": "Hypersolid",
    "url": "https://getvda.ai"
  },
  "version": "0.1.0",
  "documentationUrl": "https://witness-dao.getvda.ai/docs",
  "iconUrl": "https://witness-dao.getvda.ai/icon.svg",
  "capabilities": {
    "streaming": false,
    "pushNotifications": false,
    "stateTransitionHistory": false
  },
  "security": [],
  "defaultInputModes": [
    "application/json",
    "text/plain"
  ],
  "defaultOutputModes": [
    "application/json",
    "text/plain"
  ],
  "skills": [
    {
      "id": "governance-readiness-scan",
      "name": "Governance evidence readiness scan",
      "description": "Read a DAO's recent governance activity and report what it could prove. RETURNS: counts by action kind and by binding confidence, plus two projected grades per action — the ceiling it could reach once anchored with verified anchors (ANCHORED_VALID x binding), and what a default deployment produces today (SIGNED_PENDING x binding, because no pinned trust material ships). PROVES: nothing. This is a read of public sources and no record exists yet. It is NOT an evidence report and must never be presented as one. PRIVACY: returns aggregates only. No wallet address is returned or retained.",
      "tags": [
        "dao",
        "governance",
        "snapshot",
        "safe",
        "evidence",
        "read-only"
      ],
      "examples": [
        "Scan ens.eth and tell me which governance actions could reach PROVEN.",
        "How well bound is the recent activity in the aave.eth Snapshot space?"
      ],
      "inputModes": [
        "application/json"
      ],
      "outputModes": [
        "application/json"
      ]
    },
    {
      "id": "verify-evidence-bundle",
      "name": "Verify a Witness-DAO proof bundle",
      "description": "Verify a sealed record offline from a self-contained proof bundle { record, chain?, didDocument?, anchors? } and return the closed verdict. Makes no network call: the bundle is the entire input. CEILING: this hosted surface can never return ANCHORED_VALID. ANCHORED_VALID requires an anchor verified against material pinned in advance — a Bitcoin block-header source, Rekor’s log key, or a trusted TSA root. This service pins nothing on your behalf, because a hosted default \"trusted root\" would reintroduce exactly the vendor trust the anchoring layer exists to remove. Anchors you supply are reported as asserted, counted separately, and do not contribute to the verdict. IMPORTANT: verification run on the vendor's server by the vendor's code is a convenience, not independent verification — it cannot answer \"what if the vendor is lying?\". Tell the human to run `npx witness-dao verify --bundle ./proof.json --did ./did.json` on their own machine, and to fetch the DID document themselves.",
      "tags": [
        "verification",
        "cryptography",
        "offline",
        "read-only",
        "no-auth"
      ],
      "examples": [
        "Verify this proof bundle and tell me whether the verdict is compliance-grade.",
        "Does this record verify as BROKEN or is the bundle just incomplete?"
      ],
      "inputModes": [
        "application/json"
      ],
      "outputModes": [
        "application/json"
      ]
    },
    {
      "id": "evidence-strength-matrix",
      "name": "Explain the evidence-strength calculus",
      "description": "Return the complete (verification state x binding confidence) matrix with the reasoning for every cell. Pure — no network, no inputs. Use this before characterising any grade to a human, so the explanation is the product's own reasoning rather than your paraphrase. Note that exactly ONE cell is PROVEN (ANCHORED_VALID x onchain_confirmed); ANCHORED_VALID x api_asserted is WEAK, because perfectly anchored hearsay proves what we were told and when, not that it happened.",
      "tags": [
        "explanation",
        "grading",
        "read-only",
        "no-auth"
      ],
      "examples": [
        "Why is an anchored record of a Snapshot tally only graded WEAK?",
        "What exactly does PROVEN require?"
      ],
      "inputModes": [
        "application/json"
      ],
      "outputModes": [
        "application/json"
      ]
    },
    {
      "id": "witness-dao-capabilities",
      "name": "What Witness-DAO does, does not do, and does not yet do",
      "description": "Return an authoritative description of this product: what ships, what is free, what is deliberately unavailable on this hosted surface, and the specific claims that must never be made about it. Call this before answering any question about what Witness-DAO can do. It exists so you do not have to infer capabilities, and so you do not repeat claims — MiCA readiness, AML screening, EU AI Act Article 12 logging, customer-held signing keys — that are false about this product.",
      "tags": [
        "metadata",
        "read-only",
        "no-auth"
      ],
      "examples": [
        "Can Witness-DAO do our MiCA reporting?",
        "Does the DAO hold its own signing key with this product?"
      ],
      "inputModes": [
        "application/json"
      ],
      "outputModes": [
        "application/json"
      ]
    }
  ],
  "supportsAuthenticatedExtendedCard": false,
  "metadata": {
    "recordSigningDid": "did:web:witness.getvda.ai",
    "recordSigningNote": "Sealed records are signed by VDA Witness under that DID. This card is served by the Witness-DAO site, which is a different service and attests no evidence.",
    "cardSigned": false,
    "cardSigningNote": "This card is not yet JWS-signed. Signing requires a service identity for this host, which will be published only alongside an explicit statement that it is not the record-signing identity above.",
    "mcpEndpoint": "https://witness-dao.getvda.ai/mcp",
    "packageForLocalUse": "witness-dao",
    "licence": "Apache-2.0",
    "notOffered": [
      "seal_governance_events — needs the caller's own Witness credential and git-held rules directory.",
      "liability_defence_report — never offered over a remote surface; it is a behavioural profile of an identified natural person.",
      "Governor / Tally on-chain ingest — needs a JSON-RPC endpoint the caller chooses, not one this service picks for them."
    ],
    "mustNeverBeClaimed": [
      "MiCA reporting or AML/sanctions screening — no such capability exists.",
      "EU AI Act Article 12 logging — a category error; Article 12 governs high-risk AI systems, this seals human governance actions.",
      "That the customer holds the signing key — there is no local signing in the codebase."
    ]
  }
}