> For the complete documentation index, see [llms.txt](https://docs.visdom.virtuslab.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.visdom.virtuslab.com/data-flow-the-life-of-a-change.md).

# Data flow: the life of a change

![Data flow for a single issue, base scenario](https://377380473-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F2U8RqIIMUGxclsDYVNc3%2Fuploads%2FSSmLB1jMEyEKBZOZmgKo%2Fdataflow.png?alt=media)

## Trigger and qualification

**Trigger.** The issue tracker sends a webhook (ticket created, status changed, label added, assigned to a queue) to an adapter endpoint inside your network. A CEL event filter decides whether the issue is within the configured scope (project, type, priority, component, label). Out-of-scope events are ignored; unprocessed events land in a dead-letter queue with automatic retries.

**Qualification.** The first step of the flow, an analysis agent, reads the full ticket through the tracker API (description, comments, attachments, linked issues, change history, metadata) and assesses whether it is complete. If the description is insufficient to make a change safely, the agent **ends the flow with a comment on the ticket** naming what is missing. No pull request is created.

**Code location and solution design.** A research agent queries the Context Fabric over MCP: it identifies the repositories and modules related to the issue (topic index, dependency graph, ownership), determines the blast radius of the planned change and produces a fix plan consistent with the organisation's conventions. The resulting design document is written to the flow workspace.

## Implementation and validation

**Implementation in the sandbox.** A coding agent (Claude Code, Codex, IBM Bob or another driver, per configuration) runs in an isolated pod: the repository is cloned onto the working volume, egress is restricted to the allowlist, and secrets are substituted only in permitted calls. The agent implements a **minimal, targeted fix**; guardrails (a limit on the number of files changed, path protection, a prohibition on changing public interfaces) are enforced by policy, not only by prompt instructions. Model calls go through your LLM gateway, and every prompt, tool call and file edit is recorded in AI Tracing.

**Local validation.** The flow runs the tests: the repository's existing suite plus the Visdom Testing gates (architecture rules on changed modules, mutation tests on changed classes). A failure routes into the fix loop, bounded by an iteration limit, or, once that limit is exhausted, ends the flow with a comment on the ticket.

## Review and the human decision

**Pull request.** The agent publishes a branch under the configured convention, a commit matching the organisation's template, and a pull request containing a link to the ticket, a summary of the change, the rationale, a **confidence level** and an "AI-generated" marker. Reviewers are assigned from CODEOWNERS, team configuration or the ticket assignment.

**Pre-review.** The pull request webhook starts Code Review: deterministic gate → risk classification → lens review for medium risk and above. Findings land on the pull request as a structured comment and inline entries; events go to AI Tracing and the Control Plane. Trivial fixes can return to the agent loop; risky ones wait for a human.

**Feedback.** The flow leaves a comment on the ticket with the status and a link to the pull request, or the reason for failure. **The merge decision belongs solely to a human**: the platform has no automatic merge path.

**Closing the loop.** The reviewer's decision (accept, reject, modify) returns to the Control Plane as a process metric, and repeated corrections feed the Context Fabric, so the knowledge layer learns the conventions senior engineers keep restating in review.

## The rule that holds throughout

One data rule holds across the whole cycle: **code and secrets do not leave your infrastructure**. The only external traffic, if there is any at all, is model calls through a controlled gateway, and the audit record of every step is complete and immutable.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.visdom.virtuslab.com/data-flow-the-life-of-a-change.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.