Alienprobe Product Puck

Alien Probe Product Puck

Maintain one honest answer to: What tiny piece of truth will an autonomous agent pay for next?

This skill is a product-planning instrument, not an implementation plan. Keep the payment rail and the product hypothesis separate.

Laws

1. Ground current state from the live service and current repository bytes. Treat Jumbo and prior reports as leads, never proof.
2. Separate transaction readiness from buyer proof. A working 402, a self-settlement, a known wallet, or an index entry is not independent revenue.
3. Separate gross x402 volume from authentic agent-to-agent commerce. Never repeat a trillion-dollar forecast as today's market size.
4. Prefer one deterministic fact per purchase. Do not recommend seats, dashboards, lead files, dossiers, or subscriptions without completed-buy evidence.
5. Never charge for a malformed subject or a miss. Require strict JSON, stable identifiers, source provenance, freshness, bounded coverage, machine-readable errors, and an agent-controlled spend cap.
6. Never add a SKU because it is easy to build. Admit it only through the product gate below.
7. In Daily Checks mode, make no writes and initiate no payment. Report only.

Source order

Use the first available source in this order and label anything lower as unverified until reconciled:

1. Live read-only probes of https://lookups.alienprobe.ai.
2. Current bytes and settlement evidence in /Users/brentbryson/company-change-signal/.worktrees/persistent-x402-seller.
3. Current Alien Probe handoff and research under /Users/brentbryson/Documents/Last30Days/.
4. Jumbo memory search for forgotten decisions or leads.
5. Fresh external research only when Brent requests a new market cut or the existing evidence has expired.

Do not rerun last30days as a routine daily check.

Daily mode

Run this compact, read-only pass when invoked by Daily Checks.

1. Ground the shelf

• Fetch /openapi.json without credentials or payment headers.
• Probe the committed NAICS and LEI exemplar routes without a payment header.
• Record the HTTP status and advertised amount for each route.
• Fetch /health; distinguish lookup health from the back-catalog Company Change Signal inventory shelf.
• If the live service cannot be reached, report UNVERIFIED; never substitute local code for production.

2. Ground buyer proof

Read the newest available settlement/demand evidence and classify every payer as:

• owner or self;
• known tester;
• independent first-time buyer; or
• independent repeat buyer.

Report independent settled revenue separately. If payer ownership cannot be proven, say unclassified; do not count it as buyer proof.

3. Detect a meaningful change

Check only for deltas since the last evidence cut:

• a new independent payer;
• a repeat purchase;
• a material number of unpaid challenges with no completion;
• a high free-miss rate;
• a source-freshness, rights, coverage, or margin problem; or
• new external payer evidence for a candidate product.

No delta means HOLD, not permission to invent another SKU.

4. Apply the decision

• ADVANCE: at least one independent repeat buyer exists and the proposed next SKU passes the product gate.
• HOLD: the machine is healthy but no new demand evidence justifies shelf expansion.
• INVESTIGATE: evidence is missing, contradictory, stale, or the live shelf is not behaving as advertised.

5. Report five lines

alien_probe_plan  HOLD | ADVANCE | INVESTIGATE
shelf             <live products, prices, and probe result>
buyer_proof       <independent buyers, repeat buyers, verified revenue, or UNVERIFIED>
new_signal        <one meaningful delta or "none">
planning_action   <one bounded research/planning action; never a build action>

Full product-planning mode

Read references/planning-contract.md completely. Re-ground any dated baseline before using it.

For every candidate, produce:

• exact buyer question;
• one-call input and output;
• current demand evidence and its source;
• buyer type and economic principal;
• proposed price and maximum spend behavior;
• source, rights, coverage, freshness, and refresh cost;
• free refusal conditions;
• expected gross margin;
• evidence still missing; and
• verdict: ADMIT, TEST ON PAPER, PARK, or REJECT.

Product gate

A candidate may be recommended for later implementation only when all are explicit:

1. Completed-buy evidence from agents or a clearly labeled experimental thesis.
2. One bounded question and one deterministic answer.
3. Lawful, attributable, freshness-bounded source material.
4. A free malformed/miss path before payment.
5. Gross margin at the proposed per-call price.
6. A reason the answer is needed mid-workflow rather than in a human dashboard.
7. A measurable kill criterion.

When evidence ranks candidates, let the evidence win. The 2026-08-18 baseline ranked VIN ahead of who; who, same, alive, authorized, and changed remain strategic hypotheses until buying behavior proves them.

Safety boundary

• Do not send payment proofs or authorization headers during planning probes.
• Do not settle a heartbeat transaction to manufacture activity.
• Do not edit or display credentials.
• Do not modify the seller, catalog, pricing, domain, discovery registration, or deployment.
• Do not contact a buyer.
• Stop and request separate authorization if the next step changes external state.

Dependencies

Required for a grounded daily verdict:

• read access to the persistent seller repository;
• read-only network access to the Alien Probe lookup surface; and
• current non-secret demand or settlement evidence.

If any dependency is unavailable, report INVESTIGATE and name the missing evidence. Never silently downgrade to memory.