Command Reference

tauri-agent-tools provides 38 commands for inspecting, interacting with, monitoring, post-mortem analysis, and diagnosing Tauri applications. DOM, screenshot, and storage inspection read state; eval can change it, and monitors install temporary instrumentation. Interaction commands are debug-only (require the dev bridge). Bridge-free diagnostics work on dead apps and release builds.

Command Summary

Categories

Visual Capture

• screenshot — capture full windows or specific DOM elements with real screen pixels
• diff — compare two screenshots with pixel-level difference metrics
• snapshot — capture screenshot + DOM tree + page state + storage in a single call

DOM Inspection

• dom — explore DOM tree structure, accessibility tree, computed styles
• eval — run arbitrary JS expressions in the webview (supports --file for loading from files)

Monitoring

• ipc-monitor — watch Tauri IPC calls with timing and filtering
• console-monitor — capture console.log/warn/error output
• rust-logs — monitor Rust tracing/log output and sidecar processes
• mutations — watch DOM mutations with attribute tracking

State Inspection

• storage — read localStorage, sessionStorage, and cookies
• page-state — URL, title, viewport, scroll position, Tauri detection
• store-inspect — inspect reactive store state (Pinia, Vue devtools, custom hooks)

Window Management

• info — window geometry, position, display server
• list-windows — enumerate windows with Tauri detection
• wait — poll for windows, DOM elements, or JS conditions

Interaction (debug-only)

• click — click, double-click, or right-click a DOM element
• type — type text into an input field through the native value setter and verify the write (supports --clear, --verify-timeout)
• scroll — scroll by pixels, to top/bottom, or scroll an element into view
• focus — focus a DOM element
• navigate — navigate to a route or URL
• select — select a dropdown value via the native value setter, or toggle a checkbox via native click(), verifying the write
• invoke — call a Tauri IPC command with JSON payload

Workflow

• probe — discover running bridges, check health, list window labels; when no bridge is discoverable and no explicit --port/--token/--pid is given, emits a complete result with target.alive: false and an actionable note instead of erroring
• capture — collect screenshot + DOM + page state + storage + console errors + Rust logs into a bundle
• check — run structured assertions against DOM state (selector exists, text matches, no console errors)

Bridge-free diagnostics (new in 0.7)

Work without the dev bridge — for release builds, dead apps, and sidecar processes.

• app-paths — resolve a Tauri 2 app's OS data/log/cache/config directories from tauri.conf.json
• config inspect — emit a structured snapshot of tauri.conf.json plus a capability/permission audit
• os-logs — tail the host OS log stream filtered to a Tauri bundle id (NDJSON envelopes)
• sidecar tap — wrap-and-run a sidecar binary, frame its stdout as NDJSON, validate against an optional JSON Schema
• sidecar replay — replay a recorded NDJSON stream to stdout or into a fresh process; --tap-format unwraps {dir,ts,line} tap wrapper rows, optionally filtered by --dir in|out
• forensics — one-shot bundle for post-crash analysis (composes the above; works on dead apps)
• logs — merge an app's scattered logs (on-disk tauri-plugin-log files + the live bridge /logs ring buffer) into one timestamp-ordered NDJSON stream, normalized to UTC; filter by --level/--source/--filter and --correlate to infer correlation ids (works with no bridge); --follow tails the live bridge via v0.8 non-draining cursor reads, falling back to drain-polling on older bridges (--interval <ms> sets the fallback poll interval, default 2000)

Bridge-extending diagnostics

Feature-detect via GET /version and degrade with a note against older bridges. Use --strict to fail when an endpoint is unavailable. process-tree --deep uses an OS walk and can work without a bridge when a PID is provided.

• process-tree — Tauri PID + registered sidecars rendered as a tree
• capabilities audit — live audit of declared Tauri capabilities, flags wildcard and over-broad scopes
• webview attach — webview inspector URL or platform hint (webview2, webkitgtk, wkwebview)
• health — uptime + webview readiness + per-sidecar liveness; exits non-zero when unhealthy

Super-command

• diagnose — best-effort. Composes forensics with live bridge data when reachable; degrades cleanly when not. Master summary.md includes a "Next steps" section pointing at the right deeper command.
• bundle — collect a shareable incident archive (merged logs + deep process-tree + app-paths + forensics, plus an optional UI capture) into one directory and a .tar.gz. Secrets (tokens, API keys, passwords, JWTs, AWS keys) and PII (emails, IPs, phone numbers, home paths — including base64-encoded ones) are redacted from text artifacts on write; JSON/NDJSON artifacts are re-serialized so they stay parseable; unredacted images are flagged as warnings; each phase degrades cleanly.

Common Patterns

JSON output

All commands that produce structured output support --json:

tauri-agent-tools info --title "My App" --json
tauri-agent-tools dom --json
tauri-agent-tools storage --json

Bridge auto-discovery

Commands that need the bridge automatically discover it via token files in /tmp/. You can override with explicit flags:

tauri-agent-tools dom --port 9876 --token abc123

Duration-based monitoring

Monitor commands accept --duration to auto-stop:

tauri-agent-tools ipc-monitor --duration 5000
tauri-agent-tools console-monitor --duration 10000 --level error
tauri-agent-tools rust-logs --duration 10000 --level warn

Automation contracts

• eval --json returns { "result": value } and awaits promises. wait and check --eval use JavaScript truthiness before serialization. Thrown expressions fail.
• Fatal errors with --json use { "error": { "code": "…", "message": "…", "hint": "…" } } on stderr and exit nonzero. Results stay on stdout, including failed assertions and image-threshold results. Monitor records are NDJSON; warnings go to stderr.
• wait requires exactly one condition; check requires at least one assertion. check --no-errors fails if observation is interrupted, entries are dropped, or cleanup fails.
• Timeouts and polling intervals require positive integers. Depths, click --wait, and capture --logs-duration allow zero. Ports must be 1–65535; image thresholds must be 0–100. Conflicting click, scroll, evaluation-source, and replay-destination options are rejected.
• probe --json identifies the selected PID, port, and window label and notes ambiguous auto-discovery. Use --pid for a specific app and --window-label for its webview.
• Console, IPC, and mutation collectors have independent 1,000-entry buffers, report overflow, and take a final sample on stop. Rust log readers (logs, rust-logs, capture) use independent v0.8 cursors, with a warning and destructive drain fallback on older bridges.
• capture preserves other artifacts when screenshot tools are unavailable and records partial, warnings, and errorCount in its manifest. An interrupted capture saves available evidence and cleans up its observer.
• diagnose --no-bridge skips all bridge enrichment, even with an explicit target. logs --follow requires a bridge and cannot be combined with --no-bridge.
• config inspect and config discovery accept JSON5/JSONC without damaging URLs or comment-like strings inside values.

Shareable bundles

bundle collects privately, redacts text and structured credentials, and checks for residual secrets before publishing its directory or archive. Redaction failures prevent publication. Missing sources produce failed phases and partial: true. Images remain unredacted and appear in warnings.

Only artifacts from the current run enter the archive. Unrelated existing output files are preserved and excluded; artifact symlinks are rejected. Archive failure leaves sanitized directory evidence and reports archive: null. --no-archive intentionally produces only the directory. Review artifacts before sharing.

Quote a complete replay command: sidecar replay recording.ndjson --to-exec "node my-sidecar.js". This value is split on spaces; shell syntax and arguments containing spaces are unsupported.