Agent Runtime Integrations

Veritas is intentionally agent-agnostic. These integration notes show where the framework fits when the runtime changes.

Claude Code

• keep the tracked repo-local config under .veritas/
• use npx @kontourai/veritas shadow run --working-tree as the explicit orchestration command; it prints agent-readable feedback by default
• use npx @kontourai/veritas apply stop-hook --tool claude-code when you want Stop-hook feedback at turn end
• use the generated git/runtime hooks when you want post-change automation instead of manual invocation

Cursor

• keep adapter and policy files in the repo, not in editor-only settings
• point Cursor rules or project instructions at the repo-local Veritas governance block
• use npx @kontourai/veritas apply stop-hook --tool cursor to generate a thin wrapper around .veritas/hooks/stop.sh
• prefer veritas report for review-only lanes and veritas shadow run for proof-plus-eval lanes

GitHub Copilot Workspace

• treat Veritas as the repo-local contract surface, not as Copilot-specific policy
• run Veritas in CI to keep the evidence path independent from the editor surface
• use the same .veritas/ artifacts for local and remote execution

Aider

• keep Veritas commands in repo scripts so Aider sessions can invoke the same surface as every other runtime
• prefer explicit report or shadow-run commands over ad hoc shell snippets
• store adapter and policy changes in normal reviewable commits

Common Rule

No matter which runtime you use, keep these pieces repo-local and reviewable:

• adapter
• policy pack
• team profile
• AI instruction governance block
• stop hook shell contract
• generated evidence and eval flow