Agentic QA for VS Code

Run the agentic-qa runtime from the editor, approve its writes in a dialog instead of a terminal prompt, and read the ledger it leaves behind.

The runtime does the work. This extension exists for the two things an editor does better than a terminal: collecting an approval properly, and making a run's evidence readable.

What it adds

An approval dialog. The runtime's --approval file mode writes each batch of gated calls to approvals/<id>.json and waits. This extension watches for that file and shows you the batch — every call, every argument — as a modal.

Two properties are deliberate and not configurable:

Dismissing the dialog denies. Escape, clicking away, closing the window: all deny.
Approve is never the default button. An approval you can collect by pressing Enter is not an approval.

If you need longer to decide, choose Open the request — the run keeps waiting until aqa.approvalTimeoutSeconds elapses, and the status bar will re-ask.

A runs view. Every run in .aqa/runs/, newest first, with its status and event count. Expand one to see its events in order; click an event to open the recorded JSON. This is the same evidence aqa replay renders, browsable rather than printed.

Commands

command	does
`Agentic QA: Run story for work item…`	prompts for a work item id, then runs with file approval
`Agentic QA: Dry run`	prints every reachable tool and its gate class; spends nothing
`Agentic QA: Replay a run as Markdown`	opens the rendered run report
`Agentic QA: Show runtime version`	reports which `aqa` was found, and where

Start with the dry run. It makes no model calls and no tool calls, and it answers "what could this thing do in my repository?" before you spend anything finding out.

Requirements

The runtime is not bundled. The extension looks for aqa in this order:

aqa.executablePath, if you set it
node_modules/.bin/aqa in the workspace folder
aqa on your PATH

If none is found it offers to run npm install -g @vijaypjavvadi/agentic-qa in a terminal, where you can see and stop it.

They are versioned separately on purpose. The ledger event shapes are a contract, and someone pinning an older runtime should not be forced to upgrade it because the editor integration changed.

You also need what the runtime needs: a .env with your credentials, an ai-quality.config.yaml, and a workspace folder with a playwright.config.ts. See USAGE.md, and ADOPTION.md if you are introducing this to a team.

Settings

setting	default	what it is
`aqa.executablePath`	(empty)	absolute path to `aqa`; empty means search
`aqa.configFile`	`ai-quality.config.yaml`	policy config, relative to the workspace folder
`aqa.workspaceDir`	`qa`	the only folder the agent may write to
`aqa.ledgerDir`	`.aqa`	ledger root; runs live in `<ledgerDir>/runs/`
`aqa.approvalTimeoutSeconds`	`1800`	silence past this counts as denied

What it does not do

It does not edit ai-quality.config.yaml. That file is your policy — what the agent may do and where — and it belongs in the repository under review like any other policy, not behind a settings form that makes it feel like a preference.

License

MIT.