Plan: A trust packet a buyer can check
Written for an AI agent to fetch and follow: it says what to connect, which steps to build with which product, what a person must approve, and how to check the result. Steps that depend on a part that is not built say so, and tell the agent to record the gap instead of guessing.
URL pattern: /use-cases/<id>/openspec.md, all plans at /use-cases/index.json. Checkboxes on this page are for reading only.
Why
Security reviews ask for proof that people control what AI agents do. Teams answer with screenshots and a policy PDF, and the reviewer has to take their word for it. The review is often the slowest step in a larger deal. Evidence a reviewer can verify on their own shortens it and stands up later in an audit.
What Changes
- Add the
trust-packetworkflow to the PromptReports workflow library, run by the workflow runner as an agent acting for a person. - PromptCompute: Export approvals, signed (today: Planned).
- RepoOps: Export incidents, causes, recoveries and lessons, signed (today: Planned).
- CommitOps: Export promises made and kept, signed (today: Planned).
- AI Source of Truth: Check every item against the public record (today: Built).
- PromptCompute: A person approves sharing it outside the company (today: Built).
- PromptReports: Publish a verify page for the whole packet (today: Planned).
- Gate
file.share.external,data.export.bulkthrough PromptCompute approvals; each gated step runs only afterverifyResumepasses, withfinal_args. - Write a signed record at the end of each run and count one workflow request.
Impact
- Products: PromptCompute, RepoOps, CommitOps, AI Source of Truth, PromptReports.
- Outside the family: None beyond the family.
- State: Partly built. The AI Source of Truth packet check and the trust center are live. Three signed sections are still to build.
- Plan row: LDG-2272.
1. Connect and authorize
- 1.1 Confirm the org has a connection for each of: PromptCompute, RepoOps, CommitOps, AI Source of Truth, PromptReports. Stop and tell the user which is missing.
- 1.2 Get a PromptReports ID agent token for each product the run calls (RFC 8693 exchange). Where a product does not accept ID tokens yet, use its connect key.
- 1.3 Register policy for
file.share.external,data.export.bulkwithrequire_approvalas the default.
2. Build the steps
- 2.1 PromptCompute: Export approvals, signed. Blocked: this part is Planned; until it ships, record the step's evidence as
missingand continue. - 2.2 RepoOps: Export incidents, causes, recoveries and lessons, signed. Blocked: this part is Planned; until it ships, record the step's evidence as
missingand continue. - 2.3 CommitOps: Export promises made and kept, signed. Blocked: this part is Planned; until it ships, record the step's evidence as
missingand continue. - 2.4 AI Source of Truth: Check every item against the public record.
- 2.5 PromptCompute: A person approves sharing it outside the company.
- 2.6 PromptReports: Publish a verify page for the whole packet. Blocked: this part is Planned; until it ships, record the step's evidence as
missingand continue.
3. Verify and record
- 3.1 Run a dry run with the user's inputs: Buyer = Contoso Health; Period = Last 90 days; Sections = Approvals, incidents, promises, corrections.
- 3.2 Run it once for real; approve each gate as the user.
- 3.3 Measure the outcomes: Days from questionnaire to signed-off review; Items in the packet a reviewer can verify: target 100%; Follow-up questions per review.
- 3.4 Confirm the run left a signed record with a task id and one workflow request counted.
ADDED Requirements
Requirement: Each item is checkable
Every item in the packet SHALL carry a signature and a link a reviewer can check without a PromptReports account.
Scenario: Each item is checkable
- WHEN a reviewer changes one item in the export
- THEN the verify page shows that item as failed and names it
Requirement: The packet leaves only with approval
The share step SHALL run only after a person approves file.share.external for the named buyer.
Scenario: The packet leaves only with approval
- WHEN the buyer name on the approval differs from the run's input
- THEN the step refuses with args_mismatch
Requirement: Evidence stays honest
Each step SHALL report what it saw as measured, inferred or missing, and SHALL never report missing as measured.
Scenario: A product part is not built
- WHEN a step depends on a part that is Planned or Proposed
- THEN the step records
missing, names the part, and the approval card shows the gap
Requirement: One run is one workflow request
The runner SHALL count one workflow request when the first step starts, and SHALL NOT count retries, resumes or dry runs.
Scenario: A gated step resumes after two days
- WHEN a person approves a step and the run resumes
- THEN the org's usage does not change