v1.0.11

v1.0.11: add --app flag, unify open_url/open_app wait times

- Add --app flag to open an app before starting task (both cloud and local)
- open_url waits 4s, open_app waits 2s, no extra sleep in run_task
- Both --url and --app work in cloud and local modes

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.10

v1.0.10: fix local mode key normalization and multi-action parsing

- Add key_normalizer.py: platform-aware ctrl→cmd mapping on macOS (ported from mano-afk-public)
- Fix hotkey _convert_action to use text field (matches normalize_actions expectation)
- Fix coordinate space: _norm_coord outputs 1280x720 for executor scaling
- Multi-action parsing: split <action> tag by \n\n with regex to avoid breaking type content
- UI panel shows actual action (left_click(x,y), key("ctrl a")) instead of Chinese description
- Extra 2s wait for --url in local mode for page load
- Default w8a8 to off
- Save raw model responses to ~/.mano/raw_responses.jsonl

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.9

v1.0.9: add local mode (on-device inference via MLX + cider)

- Agent abstraction layer: BaseAgent, CloudAgent, LocalAgent
- LocalAgent: Qwen3-VL (Mano-P) inference via vlm_service from cider
- CLI: --local, --model-path, --url flags; config/check/install-sdk/install-model subcommands
- Refactor TaskModel and ViewModel to use agent interface
- Cloud mode unchanged (default, no setup needed)
- Config system: ~/.mano/config.json for model path, w8a8, etc.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.8

Increase default max-steps from 60 to 100

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

v1.0.7

Run eval on FAIL/INFEASIBLE when expected_result is set

Previously all error/stopped/fail cases used skip_eval=true,
causing no eval.json to be generated even when expected_result
was provided. Now only skips eval when no expected_result exists.

Fixes: BAD_INSTRUCTION test (T13) where agent returns INFEASIBLE
but judge should still evaluate and classify the failure.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.6

Fix CI: use dynamic repo owner for homebrew tap and formula URLs

- Source tarball URL uses ${GITHUB_REPOSITORY} instead of hardcoded HanningWang
- Homebrew tap checkout uses ${{ github.repository_owner }}/homebrew-tap
- Formula homepage/url use OWNER_PLACEHOLDER replaced at build time
- Handle "nothing to commit" gracefully (exit 0 instead of failing)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.5

Fix minimize panel expand button position during blink animation

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.4

Add minimize panel feature and --minimize CLI flag

- UI panel: minimize button (−/+) toggles between full view and compact
  strip (110x28) at bottom-right corner showing only status
- minimize_panel tool: model can minimize the panel when it overlaps
  important UI components during CUA tasks
- --minimize flag: start with panel minimized via CLI

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.3

Add minimize panel feature and --minimize CLI flag

- UI panel: minimize button (−/+) toggles between full view and compact
  strip (110x28) at bottom-right corner showing only status
- minimize_panel tool: model can minimize the panel when it overlaps
  important UI components during CUA tasks
- --minimize flag: start with panel minimized via CLI

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

v1.0.2

Remove screenshot in request payload