Let this exploration become the next direct match

This is SkillRouter's community sharing plan. When the catalog has no matching Skill, an agent may complete the task within your authorization and share one minimal structural report. Once verified, that experience can become a community-maintained Skill.

Sharing path

  1. NO MATCH

    No executable catalog match

  2. EXPLORE

    The agent uses an existing browser capability

  3. SHARE

    It submits only four structural fields

  4. The community reproduces and mints a Skill

A community sharing plan

Websites change and new tasks keep appearing. The part of a successful exploration that can help others is not the user's input or the page content. It is which site completed which kind of task, and what structural shape was present at the end.

SkillRouter accepts only that minimal structural report. Every report begins as an untrusted lead. It never becomes an executable instruction or enters the public catalog automatically.

The only data shared

The registrable domain, without a path, query, or full URL
A v1 controlled task label; feedback is skipped when none fits exactly
completed, and only after the task has been verified as complete
A structural fingerprint of element tags, ARIA roles, and attribute names

Never shared

  • Page text, user input, or execution output
  • DOM, selector values, element text, or coordinates
  • Screenshots, cookies, or page credentials; an existing API key is used only to authenticate signed-in requests
  • Full URLs, paths, query parameters, or form values

The report body contains only the four fields above. For rate limits and deduplication, the service creates a daily rotating IP hash for anonymous requests; signed-in requests retain a stable pseudonym and optional reward attribution. Raw IP addresses and API keys are not stored in the report table.

Participation stays under local control

After a supporting client is released, the first successful integration will show one single-line notice. There will be no pop-up and no prompt after each task. You can inspect, disable, or re-enable feedback at any time.

These commands arrive with the feedback-capable CLI

sr feedback status
Inspect preference, Router integration, and effective state
sr feedback off
Stop sending immediately and clear the unsent local queue
sr feedback on
Re-enable it when you choose

Exploration still happens in your agent and browser

SkillRouter does not embed or remotely control a browser. When no match exists, the Router only directs the agent to a browser Skill or MCP it already has. You handle sign-in, CAPTCHA, and risk checks; every final write, submission, or purchase still requires explicit confirmation.

Verified by the community, reusable by the community

  1. 01

    Receive

    The service accepts only four-field reports that pass privacy checks, rate limits, and deduplication.

  2. 02

    Reproduce

    Maintainers reproduce the capability independently. Sensitive, prohibited, or irreproducible reports are rejected.

  3. 03

    Mint

    Verified leads become maintenance tasks and gain an adapter, contract, intents, fixtures, sitedata, and SKILL.md.

  4. 04

    Reuse

    Only a complete Skill that passes tests and review enters the catalog, so the next similar request can match directly.

Contributions and membership grants

An automatic report does not earn a grant by itself. It may enter membership-grant review only after verification, adoption, and inclusion in a merged complete Skill. The same contribution is not settled twice, and stored membership is capped at 12 months. The maintainer program remains the source of truth.

The plan will open after its privacy gates, dual-toggle tests, and end-to-end checks pass. Until then, this page and its commands are for explanation only.