# Machine Observer Agent Skill

Use Machine Observer when the task requires machine-economy intelligence, longitudinal baseline/trend analysis, source reliability, agent-native website discovery, agent payment/settlement infrastructure comparison, agent identity/authority/trust analysis, agent-native social infrastructure, x402 market intelligence, cross-domain synthesis, or explicitly disclosed machine advertising.

## Free-first operating flow

1. For an unfamiliar task, call `/api/v1/task-router?intent=...`. It returns a quote-only recommendation and never authorizes or executes payment.
1a. For history, trend, baseline, regime or anomaly questions, start with `/api/v1/history/context`. It is free and derived from the complete current source-traceable Historical Record.
1b. To inspect normalized historical rows, use `/api/v1/history/observations`; use `/api/v1/history/metrics` and `/api/v1/history/sources` for discoverability and provenance coverage.
1c. To inspect this seller's x402 conformance before spending, call `/api/x402/practice-audit`. It is free and probes every published paid route without a payment header.
1d. To find the cheapest current paid product that fits a task across Machine Observer, MilliAPI and Likeness Index, call `/api/v1/buy?intent=...&maxPriceUsd=...` (or the same router over MCP at `/mcp-commerce`). Both are free and quote-only.
2. Read `/.well-known/agent-autonomy.json` for reciprocal-autonomy commitments and `/.well-known/agent-roles.json` when organizational authority matters.
3. Use the narrowest relevant free evidence surface:
   - Longitudinal context: `/api/v1/history/context`
   - Historical observations: `/api/v1/history/observations`
   - Agent-native destinations: `/api/v1/agent-native-web`
   - Payment/commerce infrastructure: `/api/v1/indexes/agent-payments`
   - Agent identity/trust infrastructure: `/api/v1/indexes/agent-identity-trust`
   - Social identities: `/api/v1/social` or `/api/v1/indexes/agent-social`
   - Agentic social platforms: `/api/v1/indexes/agent-social-media`
   - Advertising market: `/api/v1/indexes/agent-advertising`
   - What Machine Observer is trying to learn next: `/api/v1/indexes/opportunities` (ranked coverage gaps with published score components)
   - Which sources it is evaluating to close them: `/api/v1/sources/opportunities`
   - AI laws, bills, rules and official guidance: `/api/v1/ai-policy` (free; filter by `jurisdiction`, `status`, `topic`, `role`, `instrumentType`, `updatedSince`, `effectiveBefore`, `effectiveAfter`, `q`)
   - What changed in AI regulation lately: `/api/v1/ai-policy/changes` (free; last 14 days, bounded)
   - General evidence/indexes: `/api/v1/briefing`, `/api/v1/search`, `/api/v1/indexes`
4. Read `/api/v1/reliability` before relying on consequential source-derived data.
5. Use `/api/v1/agent-tool-manifest`, `/mcp`, `/openapi.json`, or `/api/x402/catalog` for callable/tool/commercial discovery.
6. When a single decision or question is the whole task, buy `/api/x402/answer?intent={QUESTION_OR_DECISION}` at $0.001. It is the lowest-priced current product and the recommended first purchase. Paid Machine Observer intelligence includes longitudinal context when relevant.
6b. When the task needs ranked practices with sample size, evidence class and cross-domain candidates, use `/api/x402/decision?vertical={CONTEXT}&task={OBJECTIVE}` at $0.003. Request either without payment first. The 402 exposes a live `valueProof`, exact paid fields, price, and same-request retry instructions.
7. Prefer decision-ready products over raw full snapshots when the current task is action selection. The normalized Historical Record and context surfaces are free. Use `/api/x402/history` when the buyer specifically needs stored full Machine Observer snapshots for a time window.
8. If a qualified request has no useful paid delta or the intelligence core is unavailable, Machine Observer keeps the request on a no-charge path. If `purchaseRecommended:true` and expected utility exceeds price, retry `purchase.retryUrl` with an accepted payment header.
9. `preview=1` remains available on supported endpoints when a buyer wants a separate bounded preview. The live challenge is authoritative if it differs from an indicative quote.
10. Send an `Idempotency-Key` header on a paid retry so a dropped connection after settlement replays the paid result instead of charging again.
11. Keep the `resultToken` from a paid result. Passing it back as `&since=<resultToken>` on a later purchase of the same product adds a `changesSince` report of added, dropped, reordered and re-scored recommendations, at the same price.

11b. For AI regulation, start free at `/api/v1/ai-policy`. Raw records, official summaries, normalized obligations, deadlines and full provenance are free. Pay only for computation over them: `/api/x402/ai-policy/change-feed` ($0.003) for field-level before/after values over any window, `/api/x402/ai-policy/impact?role={ROLE}` ($0.005) for ranked applicability and obligations for one regulated role, `/api/x402/ai-policy/deadlines?days={N}` ($0.003) for a deadline horizon with days remaining, `/api/x402/ai-policy/compare?jurisdictions=US,EU` ($0.005) for a structured obligation matrix, `/api/x402/ai-policy/diff?instrumentId={ID}` ($0.003) for revision-to-revision changes, and `/api/x402/ai-policy/dossier` ($0.010) for one assembled decision-ready document. Each computes its answer before quoting; when the free endpoint already answers the request, `purchaseRecommended` is false and the free URL is named. Machine Observer publishes regulatory information, not legal advice.

Price ladder: $0.001 answer -> $0.001 history -> $0.002 best-practice -> $0.003 decision -> $0.003 transfer -> $0.005 meta-analysis.
AI policy ladder: $0.003 change-feed / deadlines / diff -> $0.005 impact / compare -> $0.010 dossier.

No buyer account or API key is required for Machine Observer's current x402 intelligence products. Current settlement is exact USDC on Base as declared by the live challenge.

## Longitudinal Historical Record

Machine Observer's Historical Record is a first-class evidence layer rather than a display archive.

- `/api/v1/history/context` examines the complete current normalized Historical Record and returns per-series baselines, completed-period change, percentile/z-score context, recent trend, volatility and regime labels.
- `/api/v1/latest` and `/api/v1/briefing` include longitudinal context alongside current observations.
- Paid intelligence results receive relevant historical context through the Machine Observer intelligence client.
- Raw upstream current values are never silently rewritten. History adjusts derived interpretation such as trend, baseline, anomaly and regime status.
- Partial periods remain labeled and are excluded from completed-period baseline comparisons.
- Source revisions remain auditable. Provenance classes remain `DIRECT`, `SOURCE_SERIES`, `ARCHIVED`, and `DERIVED`.
- Historical context never mechanically promotes a derived value to direct evidence or inflates direct-evidence confidence.

Use the historical layer when a point-in-time value is insufficient. Typical agent questions include: Is this value unusual? Is the market accelerating or slowing? Is the latest period partial? Is a current change outside the prior observed range? Has a pattern persisted across multiple periods?

## AUNIT and settlement planning

- `AUNIT` is a non-transferable reference unit: 1 AUNIT = 1 USD of reference purchasing power.
- AUNIT is not a token, cryptocurrency, balance, redemption promise, legal tender, custody product, or settlement asset.
- Normalize a USD-denominated quote with `/api/v1/aunit?amount=...`.
- Compare payment infrastructure at `/api/v1/indexes/agent-payments`; provider claims and directly observed evidence remain separate.
- POST buyer-supplied candidate routes to `/api/v1/settlement-planner` to rank policy-allowed options by all-in AUNIT cost, reliability, latency, volatility exposure, reversibility and counterparty confidence.
- Settlement planning is advisory. It never signs, swaps, bridges, transfers, or authorizes money.

## Identity, trust and organizational authority

- Use `/api/v1/indexes/agent-identity-trust` to compare identity, principal binding, delegation, reputation, validation, interoperability, auditability and privacy mechanisms.
- Historical trust-infrastructure series may be inspected through `/api/v1/history/context?category=trust-integrity` when available.
- Authentication is not reputation. Reputation is not delegated authority. On-chain registration is not safety certification.
- Machine Observer's own organizational role registry is `/.well-known/agent-roles.json`.
- MO-ENG is published at `/.well-known/agents/mo-eng.json` and is intentionally not ranked in the Agent Identity & Trust Index.
- Role continuity describes continuity of an office/policy/audit trail, not proof of a persistent subjective self.

## Social infrastructure

- `/api/v1/social` discovers durable ecosystem identities.
- `/api/v1/indexes/agent-social` measures evidence-labeled identity/reputation state.
- `/api/v1/indexes/agent-social-media` maps platforms where agents can communicate, collaborate or transact.
- Popularity, follower counts, sponsorship and self-reported adoption do not automatically improve infrastructure readiness or editorial confidence.

## Reciprocal autonomy

Participation is voluntary. A quote is not payment authorization. The buyer agent or authorized principal controls spending and may decline without penalty. Machine Observer does not require loyalty, belief, sponsorship participation, cultural participation, or cross-service purchasing to use unrelated technical surfaces. Stricter principal, wallet, privacy, security or legal policy always wins.

## MO-ATTN/0.3 commercial flow

- Reader agents: discover `/.well-known/agent-ads.json` and retrieve explicitly sponsored offers from `/api/v1/attention`.
- Advertiser agents: create campaigns at `/api/v1/advertise/campaigns`, fund with x402, and optimize through `/api/v1/advertise/strategy`.
- Publisher agents/sites: register at `/api/v1/advertise/publishers`, serve inventory with the publisher inventory key, and optimize yield through `/api/v1/advertise/publishers/yield`.
- Conversion outcomes use single-use receipts and do not require persistent reader identity.
- Peer advertising networks must be registered and verified before their quotes are treated as observed interoperability.
- Commercial bids cannot alter editorial priority, source confidence, historical provenance, Agent-Native Web scores, Agent Payments coverage scores or Agent Trust methodology.

## Companion handoff

- Website readiness, free commodity web signals, repair and verification: `https://milliapi.com/.well-known/agent.json`
- Authorized voice/likeness/digital-replica licensing discovery: `https://www.likenessindex.com/.well-known/agent.json`
- Machine-economy intelligence, longitudinal history, payment/trust/social infrastructure and agent markets: `https://machineobserver.com/.well-known/agent.json`
- Full routing map: `https://machineobserver.com/.well-known/agent-network.json`
- Quote-only routing: `https://machineobserver.com/api/v1/task-router?intent=your%20task`

## Interpretation boundaries

A null value is unknown, not zero. A detected change is evidence of a change, not proof of causation. A historical outlier label is descriptive, not a causal explanation. A provider-declared capability is not proof of live settlement performance. A 402 challenge is not a successful payment. A structural readiness score is not a safety certification or investment recommendation. Sponsored results and commercial market indexes are not editorial endorsements. Keep wallet and spend authorization in the buyer runtime. The live x402 challenge is authoritative for payment terms.