sethjuarez/auditaur

Auditaur debug workflow

Use Auditaur as the first diagnostic surface for Tauri app debugging. Prefer machine-readable JSON when acting as an agent. The canonical mental model is: Auditaur observes the app; it does not replace the app's normal dev command.

Intent map

Translate common user phrases to the appropriate Auditaur workflow:

User phrase	Agent interpretation
"observe the app"	Attach to the already-running app and watch readiness.
"start with Auditaur" / "Auditaur start the app"	Run auditaur start when .auditaur/config.json exists; otherwise start the normal app command under Auditaur observation with debug run -- <normal command>.
"debug the app with Auditaur"	Check readiness first, then inspect logs/timeline/traces/IPC.
"drive the app" / "click in the app"	Require the Tauri-native drive bridge, then use auditaur drive.
"run a drill" / "smoke test the app"	Run auditaur drill for the configured default drill, or auditaur drill <name> for a named drill.
"manual approval" / "human gate" / "OAuth consent"	Use a drill script gate so the app session stays pinned while the human action completes.

Preserve the app's normal startup command. Prefer the simple repo-configured flow when available:

auditaur start
auditaur drill
auditaur inspect
auditaur stop

These commands read .auditaur/config.json and share .auditaur/session.json by default so agents do not need custom wrappers for readiness polling, session IDs, database paths, or process cleanup.

Use auditaur start --json when an agent needs machine-readable startup output. It emits one final JSON object with the exact app session, process, selectors, session file path, and any generated named ports. auditaur stop stops the recorded process tree and removes the session file after cleanup so follow-up commands do not target a dead session.

Readiness first

Before reading logs or driving the UI, establish what is ready:

auditaur debug --app <app-name> --json status

For a running app, watch until core telemetry is ready:

auditaur debug --app <app-name> --active --json watch --until-ready --timeout-seconds 120

If the task requires frontend telemetry, add --require-frontend. If the task requires WebView selector actions, use the Auditaur in-app drive bridge: enable initAuditaur({ driveBridge: true }) in exactly one debug/test WebView per Auditaur session and add --require-drive-bridge to readiness watches.

npm run tauri dev
auditaur debug --app <app-name> --active --require-frontend --json watch --until-ready --timeout-seconds 120

On Windows PowerShell:

npm run tauri dev
auditaur debug --app <app-name> --active --require-frontend --json watch --until-ready --timeout-seconds 120

Auditaur drive uses the Tauri-native in-app drive bridge. The app must explicitly enable initAuditaur({ driveBridge: true }) in exactly one debug/test WebView per Auditaur session, then auditaur drive sends bounded requests through that bridge.

Starting the app

Prefer attach mode by default: let the developer, IDE, Tauri dev server, or existing terminal own app startup, then use auditaur debug watch to observe readiness. This preserves the user's normal environment, debugger, hot reload, and terminal output.

Use wrapper mode only when the agent or a smoke script needs to own a repeatable run. Wrapper mode should still start the app through its normal command; Auditaur observes that process instead of replacing the app startup system.

Scenario	Preferred mode
Human local debugging	Attach to the already-running app
IDE/debugger/Tauri dev workflow is already running	Attach
Agent needs an end-to-end validation run	Wrapper
Dogfood or CI-like local smoke pass	Wrapper

If the agent should start the app, wrap the existing command:

auditaur start

auditaur start uses .auditaur/config.json, reports readiness, writes .auditaur/session.json, and intentionally leaves the app running after it becomes ready. If the config declares named ports, start reserves random local ports, expands {{port:name}} placeholders in the command, sets configured port environment variables, and records the chosen ports in the session file and JSON output. If no config exists yet, use the lower-level form:

auditaur debug --app <app-name> --require-frontend --json run --timeout-seconds 180 --write-session .auditaur/session.json -- npm run tauri dev

debug run reports readiness and intentionally leaves the app running after it becomes ready. When --app is supplied, it ignores matching discovery records that existed before spawn, waits for the new Auditaur session, and pins readiness to that exact session/database/pid. Use --write-session <path> for agent-owned startup so later commands can read sessionId, instanceId, pid, databasePath, and the generated selector argument arrays instead of guessing with --active or --latest. Clean up the spawned app process with auditaur stop when the validation is done.

Interpreting readiness

Inspect the stages array:

• app_discovery: Auditaur discovery file exists.
• heartbeat: app heartbeat is fresh.
• telemetry_database: SQLite database exists and schema validates.
• session: a session row is queryable.
• window: Tauri window telemetry exists.
• backend_telemetry: backend/plugin logs, spans, or window rows exist.
• frontend_telemetry: frontend logs/errors/IPC/events exist; required only with --require-frontend.
• drive_bridge: Tauri-native in-app drive bridge status exists and has a fresh heartbeat; required only with --require-drive-bridge.
• cdp_endpoint: legacy browser-debug endpoint readiness; checked only when explicitly requested.

If readiness is false, follow the stage messages and hints instead of guessing from terminal output.

Driving the WebView

Inspect the Tauri-native drive bridge before mutating the app:

auditaur drive --app <app-name> --active --json inspect

If target selection is needed, use the bridge target:

auditaur drive --app <app-name> --active --json click --target auditaur-bridge --selector '<css>'

Use read-only actions first when possible:

auditaur drive --app <app-name> --active --json exists --selector '<css>'
auditaur drive --app <app-name> --active --json text --selector '<css>'
auditaur drive --app <app-name> --active --json snapshot --selector body --output failure.json
auditaur drive --app <app-name> --active --json screenshot --selector body --output failure.png --snapshot-output failure.json

Review snapshot artifacts before sharing them; they may contain DOM text, URLs, or other sensitive content.

For text entry, use fill when a direct DOM value setter plus input/change events is enough. Use type when a framework-controlled input or textarea needs focused text insertion:

auditaur drive --app <app-name> --active --json type --selector 'textarea' --value 'hello' --visible-only

Prefer --visible-only (or --visible) with selector actions (wait, exists, text, click, fill, type, hover, select, check, and uncheck) when validating modals, focus overlays, or fullscreen shells that leave duplicate hidden DOM behind.

If drive inspect reports bridge.status inactive, enable driveBridge in the frontend. Once active, use wait, exists, text, click, fill, type, press, hover, select, check, uncheck, evaluate, snapshot, and screenshot through the Tauri-native bridge. Bridge screenshots first try native WebView capture (screenshotBackend=tauri_native_webview_snapshot) for occlusion-free WebView pixels; selector screenshots crop that WebView image and report screenshotScope=selector plus selectorRect. If WebView capture fails, Auditaur falls back to native window capture (screenshotBackend=tauri_native_window_xcap) and then to the DOM text summary PNG (screenshotBackend=bridge_dom_summary_canvas) with error metadata. evaluate runs arbitrary JavaScript in the WebView, so keep the bridge restricted to development/test sessions. If the bridge is inactive, use Auditaur telemetry (timeline, ipc, traces, explain) for truth and pair it with Accessibility automation only when manual UI input must be simulated.

Legacy compatibility: --cdp-port may appear in older commands, but auditaur drive ignores it. Use positive wording for drive commands: say they use the Tauri-native drive bridge.

Inspecting telemetry

After readiness, use structured read commands:

auditaur apps --json
auditaur sessions --json
auditaur logs --json
auditaur errors --json
auditaur ipc --json
auditaur events --json
auditaur traces --json
auditaur timeline --json
auditaur explain --json

Prefer --session <id>, --db <path>, --active, --latest, --pid, or --instance-id when there are stale or multiple sessions.

Human gates in drills

Use drill human gates for workflows that require real human action while preserving the same pinned app session: OAuth/device-code approval, OS permission prompts, installer elevation, external browser handoffs, camera/mic permissions, hardware security keys, or any flow where an agent should pause instead of simulating trust-sensitive input.

Add gates in a drill script and run the drill with --script <path>:

{
  "gates": [
    {
      "name": "Approve GitHub sign-in",
      "instructions": "Complete the GitHub device-code approval in the app or browser, then return here.",
      "selector": "#status-card",
      "expectText": "Signed in",
      "manualContinue": true,
      "timeoutMs": 300000,
      "choices": [
        { "id": "done", "label": "Done", "outcome": "continue" },
        { "id": "blocked", "label": "Blocked", "outcome": "fail" }
      ]
    }
  ]
}

Human gates run after Auditaur has spawned and pinned the new app session and reached readiness, but before selector/text/error/IPC/explain evidence collection. Do not restart the app or switch to a different session during the gate; the point is to keep pre-gate and post-gate evidence attached to the same session/database/pid. Gates can be selector-only monitored waits, manual continues, choices with continue/retry/skip/fail/abort outcomes, inputs, and clipboard hints. Sensitive inputs and clipboard values are redacted in drill reports by default, but local temporary gate response JSON can briefly contain raw input values until the drill consumes and removes it.

When a gate requires a manual response, Auditaur also publishes a session-local request beside the pinned session database. Agents can list and answer these through MCP:

auditaur mcp

Use list_pending_human_gates to find pending gates and respond_human_gate with the requestId, optional choiceId, optional inputs, and a clear responder value. Run the MCP server with the same Auditaur data directory as the drill so it can see the session-local request files. If the Copilot canvas extension is installed with auditaur init extension, open the auditaur-human-gate canvas for the pending gate so the user gets an action card and can click a response. Terminal ENTER/choice input remains the fallback when no MCP or canvas client is available.

Common failure patterns

• No discovery file: app is still compiling, has not launched, or the Auditaur plugin is not registered.
• Database not readable/schema invalid: app initialized partially or data directory is wrong.
• No frontend telemetry: frontend API did not initialize, no frontend action has fired, or export failed in the UI.
• Drive bridge inactive: enable initAuditaur({ driveBridge: true }) in exactly one debug/test WebView, otherwise use telemetry plus Accessibility fallback.
• Pending human gate: do not bypass it with synthetic app state. Surface the gate instructions to the user, respond through the canvas/MCP/terminal path, then continue evidence collection on the same pinned session.
• Target ambiguity: use --target auditaur-bridge; if multiple sessions match, disambiguate with --session-id, --instance-id, --pid, --latest, or --active.
• JSON output contamination: use debug run --json from a current Auditaur CLI; child startup output should not appear in JSON lines.

Validation loop

For high-confidence changes:

1. Run the relevant tests/builds.
2. Launch or attach with auditaur debug.
3. Run the configured auditaur drill; if it publishes a human gate, surface it through the canvas/MCP/terminal response path instead of replacing the manual action.
4. Drive a representative UI path through the Tauri-native bridge when UI coverage is needed.
5. Re-check debug status --require-frontend when frontend telemetry matters.
6. Inspect timeline and explain.
7. Clean up spawned app processes and temporary telemetry directories.