PPromptReports
How it works

The code loop, step by step.

Two loops, one contract. The code loop goes from a production error to a lesson the next agent reads; the revenue loop goes from a forecast to a promise kept. Each step records what it knows, and says plainly what it does not.

MeasuredInferredMissing, with a reasonWatching
Step 1

The error

An error arrives from Sentry, from browser reports, or as a route that got slower. A source map points at the file; when there is no stack trace, the route can still name it.

On repoops.ai →
Measuredno-source-map
Step 2

The deploy

RepoOps records which deploy served the error and the history before it. Changes outside git count too: feature flags, migrations, Railway changes, Dependabot and provider status pages.

On repoops.ai →
Measured from the deploy recordno-deploy-record
Step 3

The change

Nine mechanisms name the commit, such as blame at the deployed version or a commit that declares its own cause. Each one's accuracy is measured by replaying past cases with only the evidence of the time.

On repoops.ai →
Blame, commit-declaredRoute filesSuspect only: first release
Step 4

The agent session

A Session-Id trailer on the commit links it exactly. Without one, time and file overlap give a strong or weak link, and when two sessions overlap RepoOps does not pick one, and the link reads weak. RepoOps reads sessions from Claude Code, Cursor, Codex and many other tools on the machine.

On repoops.ai →
ExactStrongWeakno-session
Step 5

The approval

A rollback or fix goes to a person, in the app or, for a team connected to PromptCompute, through PromptCompute, with the cause's measured accuracy attached. A denial that says the named change was not the cause becomes a label that corrects attribution.

On repoops.ai →
Approved, named approverDenied, with a reason
Step 6

The recovery

The fix is measured against the success criteria written on the task. Each criterion reads met, not met or not measured. Then RepoOps watches for 14 days. The fix reads as worked only if every criterion is met and the same cause has not come back. That is an observation, not proof the fix caused it. The 30-day rework rate sits beside it.

On repoops.ai →
MetNot metNot measuredWatching
Step 7

The lesson

A bad outcome starts a lesson candidate, and a person reviews it before it reaches the CLAUDE.md and AGENTS.md blocks the next agent reads. RepoOps replays it against that repo's past failures and shows the result, and a lesson needs that replay before it can go to the other products.

On repoops.ai →
In reviewPublishedRevoked, kept in history

One task record ties it together.

Part of the recordWhat it holdsState
Task idA T- id and an append-only list of events. Nothing is edited in place.Built
ContextA context packet: every file, brain file and lesson the agent was shown, by content digest, with one hash over the whole.Built
Decision and approvalThe cause, its evidence band, and the signed approval with the approver's name.Built
OutcomeEach success criterion and the verdict.Built
Timeline pageThe whole story on one page, each link marked measured, inferred or missing.Next
Read the full post →See the products
The revenue loop

The same steps for the revenue team.

Sales as Code, PromptCompute and CommitOps run the contract for go-to-market work, and CommitOps joins it to the code loop through RepoOps.

Forecast

Outreach drafted and a forecast kept, with the evidence behind each.

Sales as CodeOn salesascode.com →

Approval

An outbound send, such as a proposal, goes to a person before it reaches a customer.

PromptComputeOn promptcompute.ai →

Promise

Every promise made to a customer, with the work that keeps it.

CommitOpsOn commitops.ai →

Delivery

The pull requests that keep a promise show it, with their change risk and agent cost.

RepoOpsRepoOps page coming

Outcome

Forecasts graded every week against what happened. A broken promise opens a case.

Sales as Code, CommitOpsOn commitops.ai →

Lesson

A release that forecasts worse is built to file a lesson in the shared brain, reviewed like any other.

RepoOpsOn repoops.ai →
LinkWhat it doesState
Approvals for revenue actionsPromptCompute's catalog has actions for sending a proposal, enrolling a contact in a sequence and approving a discount. Today Sales as Code routes outbound sends, proposal sends among them, for a connected org.In the catalog
Promises on the change pageA pull request's change page shows the customer promises it satisfies, from CommitOps.Built, switching on
Agent cost back to CommitOpsEach hour RepoOps sends the measured agent spend and roadmap items of merged pull requests.Built, switching on
Change risk for CommitOpsCommitOps reads a pull request's risk level as plain fields.Built, switching on
A broken promise opens a caseA failed CommitOps claim becomes a case in RepoOps, so it is traced, approved and measured like a production error.Built, switching on
Read the revenue loop post →