hermes-agent/tests/agent
Teknium dad10a78d0 feat(computer-use): cua-driver backend, universal any-model schema
Background macOS desktop control via cua-driver MCP — does NOT steal the
user's cursor or keyboard focus, works with any tool-capable model.

Replaces the Anthropic-native `computer_20251124` approach from the
abandoned #4562 with a generic OpenAI function-calling schema plus SOM
(set-of-mark) captures so Claude, GPT, Gemini, and open models can all
drive the desktop via numbered element indices.

- `tools/computer_use/` package — swappable ComputerUseBackend ABC +
  CuaDriverBackend (stdio MCP client to trycua/cua's cua-driver binary).
- Universal `computer_use` tool with one schema for all providers.
  Actions: capture (som/vision/ax), click, double_click, right_click,
  middle_click, drag, scroll, type, key, wait, list_apps, focus_app.
- Multimodal tool-result envelope (`_multimodal=True`, OpenAI-style
  `content: [text, image_url]` parts) that flows through
  handle_function_call into the tool message. Anthropic adapter converts
  into native `tool_result` image blocks; OpenAI-compatible providers
  get the parts list directly.
- Image eviction in convert_messages_to_anthropic: only the 3 most
  recent screenshots carry real image data; older ones become text
  placeholders to cap per-turn token cost.
- Context compressor image pruning: old multimodal tool results have
  their image parts stripped instead of being skipped.
- Image-aware token estimation: each image counts as a flat 1500 tokens
  instead of its base64 char length (~1MB would have registered as
  ~250K tokens before).
- COMPUTER_USE_GUIDANCE system-prompt block — injected when the toolset
  is active.
- Session DB persistence strips base64 from multimodal tool messages.
- Trajectory saver normalises multimodal messages to text-only.
- `hermes tools` post-setup installs cua-driver via the upstream script
  and prints permission-grant instructions.
- CLI approval callback wired so destructive computer_use actions go
  through the same prompt_toolkit approval dialog as terminal commands.
- Hard safety guards at the tool level: blocked type patterns
  (curl|bash, sudo rm -rf, fork bomb), blocked key combos (empty trash,
  force delete, lock screen, log out).
- Skill `apple/macos-computer-use/SKILL.md` — universal (model-agnostic)
  workflow guide.
- Docs: `user-guide/features/computer-use.md` plus reference catalog
  entries.

44 new tests in tests/tools/test_computer_use.py covering schema
shape (universal, not Anthropic-native), dispatch routing, safety
guards, multimodal envelope, Anthropic adapter conversion, screenshot
eviction, context compressor pruning, image-aware token estimation,
run_agent helpers, and universality guarantees.

469/469 pass across tests/tools/test_computer_use.py + the affected
agent/ test suites.

- `model_tools.py` provider-gating: the tool is available to every
  provider. Providers without multi-part tool message support will see
  text-only tool results (graceful degradation via `text_summary`).
- Anthropic server-side `clear_tool_uses_20250919` — deferred;
  client-side eviction + compressor pruning cover the same cost ceiling
  without a beta header.

- macOS only. cua-driver uses private SkyLight SPIs
  (SLEventPostToPid, SLPSPostEventRecordTo,
  _AXObserverAddNotificationAndCheckRemote) that can break on any macOS
  update. Pin with HERMES_CUA_DRIVER_VERSION.
- Requires Accessibility + Screen Recording permissions — the post-setup
  prints the Settings path.

Supersedes PR #4562 (pyautogui/Quartz foreground backend, Anthropic-
native schema). Credit @0xbyt4 for the original #3816 groundwork whose
context/eviction/token design is preserved here in generic form.
2026-04-28 01:46:36 -07:00
..
transports fix(agent): preserve Codex message items for replay 2026-04-25 18:22:06 -07:00
__init__.py test: add unit tests for 8 modules (batch 2) 2026-02-26 13:54:20 +03:00
test_anthropic_adapter.py refactor: remove _nr_to_assistant_message shim + fix flush_memories guard 2026-04-23 02:30:05 -07:00
test_anthropic_keychain.py fix: re-auth on stale OAuth token; read Claude Code credentials from macOS Keychain 2026-04-24 07:14:00 -07:00
test_auxiliary_client.py fix(providers/gmi): post-salvage review fixes 2026-04-27 11:17:59 -07:00
test_auxiliary_client_anthropic_custom.py fix(anthropic): complete third-party Anthropic-compatible provider support (#12846) 2026-04-19 22:43:09 -07:00
test_auxiliary_config_bridge.py fix: remove legacy compression.summary_* config and env var fallbacks (#8992) 2026-04-13 04:59:26 -07:00
test_auxiliary_main_first.py fix(copilot): send vision header for Copilot vision requests 2026-04-27 08:35:50 -07:00
test_auxiliary_named_custom_providers.py refactor(memory): remove flush_memories entirely (#15696) 2026-04-25 08:21:14 -07:00
test_bedrock_1m_context.py fix(bedrock): send context-1m-2025-08-07 beta so Opus 4.6/4.7 get 1M context (#16793) 2026-04-27 20:41:36 -07:00
test_bedrock_adapter.py fix(bedrock): evict cached boto3 client on stale-connection errors 2026-04-24 07:26:07 -07:00
test_bedrock_integration.py fix(agent): handle aws_sdk auth type in resolve_provider_client 2026-04-24 07:26:07 -07:00
test_codex_cloudflare_headers.py fix(codex): pin correct Cloudflare headers and extend to auxiliary client 2026-04-19 11:59:25 -07:00
test_compress_focus.py fix: resolve CI test failures — add missing functions, fix stale tests (#9483) 2026-04-14 01:43:45 -07:00
test_compressor_image_tokens.py feat(image-input): native multimodal routing based on model vision capability (#16506) 2026-04-27 06:27:59 -07:00
test_context_compressor.py fix(compression): notify users when configured aux model fails even if main-model fallback recovers (#16775) 2026-04-27 20:08:23 -07:00
test_context_engine.py feat: wire context engine plugin slot into agent and plugin system 2026-04-10 19:15:50 -07:00
test_context_references.py fix(agent): fall back when rg is blocked for @folder references 2026-04-20 01:56:41 -07:00
test_copilot_acp_client.py fix: set HOME for Copilot ACP subprocesses 2026-04-24 05:09:08 -07:00
test_credential_pool.py fix(credential_pool): add Nous OAuth cross-process auth-store sync 2026-04-24 05:20:05 -07:00
test_credential_pool_routing.py refactor: remove smart_model_routing feature (#12732) 2026-04-19 18:12:55 -07:00
test_crossloop_client_cache.py refactor(tests): re-architect tests + fix CI failures (#5946) 2026-04-07 17:19:07 -07:00
test_direct_provider_url_detection.py fix: restrict provider URL detection to exact hostname matches 2026-04-20 22:14:29 -07:00
test_display.py fix(display): render <missing old_text> in memory previews instead of empty quotes (#12852) 2026-04-19 22:45:47 -07:00
test_display_emoji.py feat(tools): centralize tool emoji metadata in registry + skin integration 2026-03-15 20:21:21 -07:00
test_error_classifier.py feat(image-input): native multimodal routing based on model vision capability (#16506) 2026-04-27 06:27:59 -07:00
test_external_skills.py feat(skills): support external skill directories via config (#3678) 2026-03-29 00:33:30 -07:00
test_gemini_cloudcode.py fix(gemini): assign unique stream indices to parallel tool calls 2026-04-20 02:10:53 -07:00
test_gemini_free_tier_gate.py feat(gemini): block free-tier keys at setup + surface guidance on 429 (#15100) 2026-04-24 04:46:17 -07:00
test_gemini_native_adapter.py fix(gemini): fail fast on missing API key + surface it in hermes dump (#15133) 2026-04-24 05:35:17 -07:00
test_gemini_schema.py fix(gemini): drop integer/number/boolean enums from tool schemas (#15082) 2026-04-24 03:40:00 -07:00
test_image_gen_registry.py feat(plugins): pluggable image_gen backends + OpenAI provider (#13799) 2026-04-21 21:30:10 -07:00
test_image_routing.py feat(image-input): native multimodal routing based on model vision capability (#16506) 2026-04-27 06:27:59 -07:00
test_insights.py test: stop testing mutable data — convert change-detectors to invariants (#13363) 2026-04-20 23:20:33 -07:00
test_kimi_coding_anthropic_thinking.py fix(kimi): don't send Anthropic thinking to api.kimi.com/coding (#13826) 2026-04-21 21:19:14 -07:00
test_local_stream_timeout.py fix(agent): recognize Tailscale CGNAT (100.64.0.0/10) as local for Ollama timeouts 2026-04-22 14:46:10 -07:00
test_memory_provider.py fix(memory): add write origin metadata 2026-04-24 14:37:55 -07:00
test_memory_user_id.py feat(hindsight): richer session-scoped retain metadata 2026-04-22 05:27:10 -07:00
test_minimax_auxiliary_url.py fix: provider/model resolution — salvage 4 PRs + MiniMax aux URL fix (#5983) 2026-04-07 22:23:28 -07:00
test_minimax_provider.py fix(tests): resolve 17 persistent CI test failures (#15084) 2026-04-24 03:46:46 -07:00
test_model_metadata.py feat(computer-use): cua-driver backend, universal any-model schema 2026-04-28 01:46:36 -07:00
test_model_metadata_local_ctx.py fix: forward auth when probing local model metadata 2026-04-20 20:51:56 -07:00
test_model_metadata_ssl.py fix(auth): honor SSL CA env vars across httpx + requests callsites 2026-04-24 03:00:33 -07:00
test_models_dev.py feat: add Step Plan provider support (salvage #6005) 2026-04-22 02:59:58 -07:00
test_moonshot_schema.py fix(kimi,mcp): Moonshot schema sanitizer + MCP schema robustness (#14805) 2026-04-23 16:11:57 -07:00
test_nous_rate_guard.py fix(nous): don't trip cross-session rate breaker on upstream-capacity 429s (#15898) 2026-04-26 04:53:42 -07:00
test_onboarding.py fix(openclaw-migration): case-preserving brand rewrite + one-time ~/.openclaw residue banner (#16327) 2026-04-26 20:57:26 -07:00
test_prompt_builder.py feat(agent): add PLATFORM_HINTS for matrix, mattermost, and feishu (#14428) 2026-04-23 12:50:22 +05:30
test_prompt_caching.py fix(prompt-caching): skip top-level cache_control on role:tool for OpenRouter 2026-03-21 16:54:43 -07:00
test_proxy_and_url_validation.py fix(agent): normalize socks:// env proxies for httpx/anthropic 2026-04-21 05:52:46 -07:00
test_rate_limit_tracker.py feat: capture provider rate limit headers and show in /usage (#6541) 2026-04-09 03:43:14 -07:00
test_redact.py feat: replace kimi-k2.5 with kimi-k2.6 on OpenRouter and Nous Portal (#13148) 2026-04-20 11:49:54 -07:00
test_shell_hooks.py feat: shell hooks — wire shell scripts as Hermes hook callbacks 2026-04-20 20:53:51 -07:00
test_shell_hooks_consent.py fix(shell_hooks): parse hooks_auto_accept as strict bool/string, not bool() (#16322) 2026-04-26 20:48:35 -07:00
test_skill_commands.py refactor(commands): drop /provider, /plan handler, and clean up slash registry (#15047) 2026-04-24 03:10:52 -07:00
test_streaming_context_scrubber.py style: trim verbose comment blocks added by previous commit 2026-04-27 12:37:33 -07:00
test_subagent_progress.py feat(delegate): orchestrator role and configurable spawn depth (default flat) 2026-04-21 14:23:45 -07:00
test_subagent_stop_hook.py feat: shell hooks — wire shell scripts as Hermes hook callbacks 2026-04-20 20:53:51 -07:00
test_subdirectory_hints.py fix(agent): catch PermissionError in subdirectory hint discovery 2026-04-09 03:10:30 -07:00
test_title_generator.py fix(title-gen): surface auxiliary failures via _emit_auxiliary_failure 2026-04-26 21:49:34 -07:00
test_unsupported_parameter_retry.py fix(auxiliary): generalize unsupported-parameter detector and harden max_tokens retry (#15633) 2026-04-25 05:50:34 -07:00
test_unsupported_temperature_retry.py refactor(memory): remove flush_memories entirely (#15696) 2026-04-25 08:21:14 -07:00
test_usage_pricing.py fix(usage): read top-level Anthropic cache fields from OAI-compatible proxies 2026-04-22 17:40:49 -07:00
test_vision_resolved_args.py fix: pass resolved args to resolve_vision_provider_client() 2026-04-16 07:45:13 -07:00