lenses / connect v0.1

Any AI can research.
One format brings
the lens into ask.

Lenses Connect is a portable contract between research agents and the Question Genome. Bring evidence and a way of noticing. ask decides how that perspective can open a better question.

Connect a lens →Read the machine contract ↗

The API is the foundation. MCP can become one convenient doorway into it.

the simple architecture

One manifest.
Three ways in.

Today

Copy or upload

Any AI returns a lens.json. The person reviews it here and chooses whether it enters their account.

API next

Agent handoff

An authorized agent creates a draft and receives a human review link. The agent never silently installs a perspective.

Adapter

MCP

An MCP server can expose the same validate, draft, inspect, and update operations. It does not become a second lens format.

01 / ask any AI to research

Start with a job,
not a platform.

Give this brief to the AI or agent you already use. Add the perspective, source collection, permissions, and intended use.

Research and build a source-grounded lens for ask.

Return only valid JSON using schemaVersion "ask.lens/v0.1". The lens must include:
- a specific name and concise summary
- the exact corpus, its coverage, and known gaps
- who built it, which AI or agent helped, the research date, and the method
- at least three distinct attention shifts
- one clarify, one test, and one move question
- at least two honest boundaries
- source records with stable IDs, public or permission-index URLs, publishers, and the role each source played

Do not simulate a person, invent private beliefs, hide disagreements, or claim verification. Prefer primary sources. Make every movement sourceIds reference a source in the source list. Ask me for the perspective, source permissions, and intended use before researching.

02 / bring back one file

Paste a lens.
See exactly what it changes.

Connect accepts one portable manifest. Validation happens before anything can enter your lens picker.

lens.jsonUpload JSON

{
  "schemaVersion": "ask.lens/v0.1",
  "name": "A product team on learning from failure",
  "summary": "Turns six years of retrospectives into a way of noticing repeated system conditions, weak signals, and learning that never became practice.",
  "kind": "team",
  "color": "#24806b",
  "corpus": {
    "description": "Forty-two permissioned product retrospectives written between 2020 and 2026.",
    "coverage": "Planning, launches, incidents, handoffs, and team communication through July 2026.",
    "gaps": [
      "No customer-support retrospectives before 2023",
      "Former teammates have not yet reviewed the interpretation"
    ]
  },
  "provenance": {
    "builtBy": "The product team",
    "builtWith": "A research agent",
    "researchedAt": "2026-08-02",
    "method": "The agent clustered recurring conditions, preserved counterexamples, linked every question move to source records, and the team reviewed the resulting boundaries."
  },
  "attention": [
    {
      "title": "Repeated conditions",
      "description": "Looks for system conditions that recur under different incidents instead of treating every failure as unique."
    },
    {
      "title": "Weak signals",
      "description": "Surfaces warnings the team noticed but did not know how to act on."
    },
    {
      "title": "Unfinished learning",
      "description": "Checks whether an insight changed a decision, process, or owner after the retrospective ended."
    }
  ],
  "movements": [
    {
      "movement": "clarify",
      "question": "Which recurring condition is this situation another example of?",
      "purpose": "Connect the present event to the team’s longer memory.",
      "sourceIds": [
        "S1"
      ]
    },
    {
      "movement": "test",
      "question": "What evidence would show that our favorite explanation is incomplete?",
      "purpose": "Protect the team from a tidy single-cause story.",
      "sourceIds": [
        "S1",
        "S2"
      ]
    },
    {
      "movement": "move",
      "question": "What would have to change in the system for this lesson to become visible in practice?",
      "purpose": "Translate reflection into owned structural change.",
      "sourceIds": [
        "S2"
      ]
    }
  ],
  "boundaries": [
    "This lens represents patterns in a bounded document collection, not every teammate’s perspective.",
    "Private source text must never be exposed to someone using the compiled lens."
  ],
  "sources": [
    {
      "id": "S1",
      "title": "Product retrospectives, 2020–2023",
      "publisher": "Permissioned team archive",
      "url": "https://example.com/private-source-index-1",
      "role": "Evidence for recurring conditions and weak signals."
    },
    {
      "id": "S2",
      "title": "Product retrospectives, 2024–2026",
      "publisher": "Permissioned team archive",
      "url": "https://example.com/private-source-index-2",
      "role": "Evidence for unfinished learning and changed practice."
    }
  ]
}

the manifest contract

Research in.
Transparent influence out.

Connect deliberately does not accept a giant hidden prompt. It accepts the material needed to inspect and govern a perspective.

corpus

What was studied, the coverage boundary, and known gaps.

provenance

Who built it, which AI helped, when research happened, and the method used.

attention[]

The specific things this perspective makes more visible.

movements[]

One grounded question for clarify, test, and move—with source references.

sources[]

A stable source index and the role each source played in the interpretation.

boundaries[]

What the lens cannot claim, where it may fail, and what remains private.

quality and trust

A connected lens must earn its influence.

01
No person simulation
“An interpretation of someone’s public work” is acceptable. “This is what they think” is not.
02
Receipts before confidence
Question moves cite the source records that earned them. Unsupported intuition stays outside the Genome.
03
Private corpus, private material
A compiled lens may influence questions without exposing the documents, journals, interviews, or team records behind it.
04
Human approval at the boundary
An agent can prepare a draft. A person decides whether it becomes available in their ask picker.
05
Version everything
Sources, method, Genome effects, corrections, and verification belong to a specific version—not silent drift.

what comes after v0.1

Learn from the manifest before multiplying the interfaces.

1. Draft API
API keys or delegated authorization, idempotent draft creation, validation errors, and human claim URLs.

2. MCP adapter
Tools for schema discovery, validation, draft creation, inspection, revision, and status.

3. Verified publishing
Claiming, correction, version signatures, steward transfer, source permissions, and public lens URLs.

Any AI can research.One format bringsthe lens into ask.

Copy or upload

Agent handoff

MCP

No person simulation

Receipts before confidence

Private corpus, private material

Human approval at the boundary

Version everything

Learn from the manifest before multiplying the interfaces.

Any AI can research.
One format brings
the lens into ask.