Run the scan

A scan ties everything together: it runs one workflow against a target repository with the model, harness, post-scripts, and rankers you choose. The engine queues it and runs every step.

Creating a new scan: pick a workflow, choose a target, and configure the run.

Create the scan

Open Scans → New scan and work down the form:

1. Choose a workflow

   Pick the workflow you built or one of the built-in workflows. The scan will run its steps in depth order.

2. Choose a target

   Toggle Remote or Local. For remote, enter a GitHub owner/repo and a commit_sha (defaults to HEAD). For local, pick a folder you've mounted. Set repo_scope to describe what to focus on (defaults to "full repository").

3. Add dependencies and configuration (optional)

   Add dependency repos to be checked out alongside the target, and a JSON configuration blob the agents can read. To give every task a cumulative second attempt, use {"repeat_runs": 2}. The second execution of each concrete task sees only that task's earlier output and returns new records to append; leave it at 1 for the cheapest first scan.

4. Pick model, provider, and harness

   Choose a model from the available-model picker for Codex or Anthropic. For OpenRouter, enter its exact model identifier. Then choose from the providers configured for the engine, a compatible harness, and a thinking effort. The form defaults Codex to the Codex harness and Anthropic/OpenRouter to Claude Code. If no provider is listed, complete AI provider setup and recreate the stack.

5. Attach skills, post-scripts, and rankers

   Select any agent skills, one or more post-scripts (the first is the primary), and one or more severity rankers plus optional scan-specific rules.

6. Launch

   Submit. If another scan is active, choose Start immediately to run concurrently or Queue to wait until active scans finish.

The normal path is pending, optional prewarming_cache, running, post_processing, then completed. A queued scan starts when the engine is idle. A temporarily rate_limited scan retries automatically without discarding completed work. See errors and re-runs.

Warning

Each workflow branch, repeat, ranking batch, and post-script consumes model calls. Start with one repeat and a narrow repo_scope, then expand after the workflow produces useful results.

Watch it run

The scan detail page shows live status, active jobs, and any errors. As findings land, they appear in the list - ranked by your ranker, enriched by your post-scripts.

Info

Each part of this form has its own page: local vs remote, dependencies, configuration, model and harness, and more under Scans.

View the results

When the scan finishes, open any finding to see its full write-up, proof of concept, and post-script output. See Scan results for the tour.