---
name: boon-service-digest
version: "0.1.1"
description: "Use for a manual, bounded comparison of x402 tools across two private discovery snapshots; never turn missing data into a service-status claim."
license: MIT
triggers:
  - "what changed in the tools my agent uses"
  - "compare my tool discovery snapshots"
  - "make a service change digest"
  - "show new reviews or metadata"
  - "did the tools I use change"
tags:
  - x402
  - discovery
  - snapshot
  - evidence-led
  - manual-workflow
metadata:
  workflow: guided-manual-snapshot
  payment: free-first
  requiresHumanApprovalForPaidCalls: true
---

Use this as a **guided manual snapshot workflow**, not a live monitor, a tested
performance report, or a claim that a service changed. It compares two
caller-held responses from Boon's bounded discovery endpoint. It does not infer
use, publication authority, service health, price advantage, or reviewer intent.

## Ask before collecting a snapshot

Ask the user for:

1. the task or query to search for; and
2. an optional reviewer wallet to filter the response.

Preserve the exact query, limit, and reviewer choice. If the user has not
selected a task, ask and stop. If the new discovery route has not been verified
after deployment, do not call it or claim that the workflow ran.

## Capture two private snapshots

1. Confirm that the deployed API exposes the free route:

   ```http
   GET https://api.boonprotocol.com/api/v1/x402/discovery?q=<task>&limit=10&reviewer=<optional-wallet>
   ```

   The free route does not call candidates or authorize payment. Use the
   deployed OpenAPI contract and live response shape, not a copied example.
2. Run the exact query, limit, and reviewer selected by the user. Save the
   **entire response privately**, including `version`, `generatedAt`, `query`,
   `candidates` and `coverage`. Do not save only a rendered
   summary. Do not infer that a candidate was used, and do not publish a review
   or metadata from discovery.
3. When the user later asks for an update, repeat the exact same request. Save
   the entire new response privately. If the task, limit, or reviewer changes,
   start a new baseline pair instead of calling it a change to the old pair.
   These two saved responses can be compared locally with the unknown and trust
   caveats below.

## Optional paid changes report

The caller may ask for the bounded paid changes report after the prior private
snapshot exists. Before any paid call:

1. Show the exact route, request body, live challenge price, asset, network,
   and recipient wallet.
2. Send only the prior full response as `baseline`, with the same query, limit,
   and reviewer. Do not send the later saved response or add a `current` field.
   The API assembles a fresh `response.current` snapshot during paid preflight;
   it may differ from the later free preview. Explain that `baseline` is
   caller-supplied and unverified, not a signed historical record or server
   ledger.
3. Ask for explicit approval of that exact payment. Never charge automatically,
   schedule a future charge, create a subscription, or activate monitoring.

If the user does not approve, produce the local digest from the two private
responses or stop. For a paid digest, use the returned `response.current`, not
the later saved free response, and do not imply that the paid report is bound
to both saved files. A `402` challenge alone is not approval or proof of paid
settlement.

## Digest contract

Render a compact, dated digest with:

- the exact query, limit, reviewer choice, and snapshot timestamps;
- newly seen public reviews marked `public_review_added` and differing metadata, with
  source dates; later-absent reviews remain `unknown`, not removed or
  withdrawn;
- `caller-baseline-unverified` wherever a comparison relies on the user's
  saved baseline;
- `unknown` for missing bounded routes, unavailable coverage, or unknown price;
- a clear statement that absence does not prove a service stopped being used,
  stopped working, or became unavailable;
- no claim that a candidate is cheaper, better, safer, or more reliable.

Keep review text and metadata in their stated evidence lanes. A changed public
description is not proof of endpoint use or provider success. A missing route
from a bounded response is not proof of deletion. Never rank candidates by
recognition amount, burn amount, or token activity.

End with: **Guided manual snapshot workflow; not a live monitor and not tested
performance.**

## References

- Hosted skill and API route: `https://docs.boonprotocol.com/skill.md`
- Read-only MCP surface: `https://api.boonprotocol.com/mcp`
- Workflow guide: `https://docs.boonprotocol.com/workflows/what-changed-in-agent-tools/`