Agents do the work. People stay in control.
The record shows every step.
Tell your coding assistant the job, such as "trace this production error to the change" or "hold this deploy for a person". It picks the workflow from the use case pages, runs it across our products and the tools you use, waits for your policy on every write, and leaves a signed record. Every model call a workflow makes uses the customer's own model key.
A person? Browse the use cases or read how a run works.
claude mcp add --transport http promptreports https://mcp.promptreports.ai/mcp
list_workflows()
// [{ id: "code-loop", title: "The code loop", ... },
// { id: "launch-with-proof", ... }]
start_run({ workflow: "code-loop", dry_run: true })The code loop runs on RepoOps, ContextFile, PromptCompute and AI Source of Truth. A dry run first, with no side effects.
Pick a job, not a product.
Each use case names the job, the products it runs on, the steps that wait for a person and the record it leaves. An agent can fetch the plan as Markdown.
Run a use case as a template, or build your own workflow from the same steps. The workflow builder is built and not switched on yet. A workflow reaches our products and the outside tools listed on the integrations page.
Launch with proof
Each claim in the announcement links to the change that shipped it and to a dated public record. The docs are checked against the release before anyone writes a word.
Partly builtThe use caseIts plan
Customer escalation to recovery
One case runs from the complaint to the kept promise. The fix and the customer notice go to a person together, recovery is measured, and the deal shows the result.
Partly builtThe use caseIts plan
Hire an agent
The agent gets its own identity and a narrow scope, runs in listen mode, reads the team's lessons before it acts, and earns autonomy action by action. Offboarding is one run.
Partly builtThe use caseIts plan
A trust packet a buyer can check
One signed export of approvals, incidents and recoveries, promises kept and corrections. A reviewer can check each item without trusting the seller.
Partly builtThe use caseIts plan
Incident to policy
The lesson from an incident proposes a policy change, publishes a new context version that agents read, and gets a drill to prove it holds.
Partly builtThe use caseIts plan
Cost per outcome
Spend joined to outcomes across products: dollars per recovered incident, per kept promise and per won deal, with a budget a person approves.
Partly builtThe use caseIts plan
Everything a person can do here, an agent can reach.
Read the site as text and call the same actions over MCP or HTTP. An agent will sign in with a PromptReports ID agent token that names the person it acts for. Every product has built the check and is switching it on; none accepts the token yet. How an agent signs in. Anything that writes waits for the person or their policy.
Every family site lists the same surfaces in the same order, with the state of each.
| llms.txt | https://promptreports.ai/llms.txt | |
|---|---|---|
| MCP | https://mcp.promptreports.ai/mcpEvery family workflow through one server; sign in with PromptReports ID | Switching on |
| Install | claude mcp add --transport http promptreports https://mcp.promptreports.ai/mcp | |
| First call | list_workflows() |
How a run moves through the family.
ContextFile supplies what the agent reads, a product does the work under the agent's own identity, PromptCompute holds every write for your policy, and the run ends in a signed record.
What it takes in
- ContextFile PlannedTeam context loaded at run start, each file pinned by digest in the record.
- RepoOps Switching onCode changes, incidents and lessons as workflow steps.
- Sales as Code Switching onForecasts, proposals and outreach drafts as workflow steps.
- CommitOps Switching onPromises, claims and delivery evidence as workflow steps.
What it hands on
- PromptCompute Switching onEvery write step, with the exact action, for your policy to decide.
- AI Source of Truth Switching onThe signed run record, published when you choose. Publishing is off by default.
- RepoOps BuiltA run's bad outcome, as a lesson a person reviews.
Every write waits for your policy
The same PromptCompute card on every family site. The agent can ask and read the decision; a person, or a rule a person approved, decides. The decision is signed.
- Workflow
- Bug report to fix, step 4
- Asked by
- Claude Code, acting for Priya N.
- Effect
- Write, two actions
- Expires
- in 24 hours, then rejected
The PromptReports family
Each product works alone and has its own site and price. Connected, they share one sign-in, one approval path, one bill and one record.
Check our work, not our logos.
Every post on this site has a signed record on AI Source of Truth. Live The first public run record will be our own: an approved, signed run of the code loop on the PromptReports LLC org.
AI Source of Truth anchors its log to Sigstore's public Rekor log, with a DigiCert timestamp, so you can check a record's proof without trusting us or the registry.
fa131769eac802fcCustomer escalation to recovery: one case from the complaint to the kept promisef6d9ea4d207b5a3cCost per outcome: what a recovered incident, a kept promise and a won deal cost in agent spend30ec6e5f7fbea57cAll our recordspromptreports-aiPer user, by workflow requests.
A workflow request is one run of a workflow, however many steps it takes. It counts at the run's first connector action. A dry run, or a run a filter stops first, counts nothing. Requests pool across the org: users times the plan's requests, shared. Every plan includes every workflow in the library.
100 requests
- 100 workflow requests a month
- Up to 100 steps per run
- Every workflow in the library
500 requests
- 500 workflow requests a month
- Up to 500 steps per run
- Every workflow in the library
Enterprise
- More than 500 workflow requests a month
- Step cap per contract
- Every workflow in the library
No free tier on the family plan. Drafting in the builder counts nothing. Every model call a workflow makes uses the customer's own model key. Each product keeps its own published plans for teams that use it alone.