futurediffusion/claudebot

futurediffusion-Profil

Letzte erfasste Quellaktivität: 19. April 2026 um 20:16

SkillsMP-Katalog aktualisiert: 17. Oktober 2026 um 00:09

gesammelte Skills: 25

GitHub-Stars: 0
GitHub-Forks: 0

README des Repositorys

README-Quelldatei: README.md

README-Quelldatei lesen

Vollständige README anzeigen

# Claudebot Workspace Workspace reorganized to keep the active system separate from older experiments. ## What To Use - **Referencia primaria de gobernanza**: [docs/PROJECT_CONSTITUTION.md](D:/IA/CODE/claudebot/docs/PROJECT_CONSTITUTION.md). - **Nombre operativo**: **Sistema local CLI-first para automatización personal en PC**. - **Prioridades**: operación local primero, CLI-first, routing pragmático. - [orchestrator/README.md](D:/IA/CODE/claudebot/orchestrator/README.md): núcleo activo de operación local CLI-first con routing pragmático. - [tools/](D:/IA/CODE/claudebot/tools): external tooling and support repos. - [.claude/](D:/IA/CODE/claudebot/.claude): local editor and agent config. - [run_browser.py](D:/IA/CODE/claudebot/run_browser.py): direct browser automation wrapper. - [run_windows.py](D:/IA/CODE/claudebot/run_windows.py): direct Windows automation wrapper. - [run_worker.py](D:/IA/CODE/claudebot/run_worker.py): full `worker-core` wrapper. - [run_agent.py](D:/IA/CODE/claudebot/run_agent.py): high-level recommended entrypoint with locked agent selection. - [skills_bridge.py](D:/IA/CODE/claudebot/skills_bridge.py): shared bridge to discover, read, and execute local skills from one CLI. - [self_model/README.md](D:/IA/CODE/claudebot/self_model/README.md): shared self-model used by Claude, Gemini, Codex, and wrappers. - [self_model_cli.py](D:/IA/CODE/claudebot/self_model_cli.py): inspect or simulate the shared self-model from the repo root. - [episodic_memory/README.md](D:/IA/CODE/claudebot/episodic_memory/README.md): shared episodic memory for concrete attempts, failures, and fixes. - [episodic_memory_cli.py](D:/IA/CODE/claudebot/episodic_memory_cli.py): inspect similar past episodes from the repo root. - [world_model/README.md](D:/IA/CODE/claudebot/world_model/README.md): shared desktop world model for live environment state. - [world_model_cli.py](D:/IA/CODE/claudebot/world_model_cli.py): inspect the current world-state slice from the repo root. - [orchestrator/docs/CAPABILITIES_MATRIX.md](D:/IA/CODE/claudebot/orchestrator/docs/CAPABILITIES_MATRIX.md): matriz de capacidades reales vs experimentales del sistema. - [orchestrator/docs/EXECUTION_HIERARCHY.md](D:/IA/CODE/claudebot/orchestrator/docs/EXECUTION_HIERARCHY.md): pirámide oficial de ejecución, escalado por niveles y fallback extremo mouse/visión. - [docs/SKILLS_CATALOG.md](D:/IA/CODE/claudebot/docs/SKILLS_CATALOG.md): catálogo de skills operativas (dominio, IO, costo, fallbacks y owner). - [docs/DEPRECATION_POLICY.md](D:/IA/CODE/claudebot/docs/DEPRECATION_POLICY.md): estados oficiales (`active`, `discouraged`, `deprecated`, `archived`) y checklist de migración. ## Qué NO es - No es AGI. - No es mouse-first. **Status: deprecated** - No es multimodelo por defecto. **Status: discouraged** ## Valor práctico hoy Este workspace aporta valor cuando te resuelve tareas diarias de forma directa y repetible: - **Limpieza**: ordenar carpetas de trabajo, mover archivos temporales y dejar resultados en rutas consistentes. - **Organización**: crear resúmenes, notas operativas y salidas en `tasks/output/` con estructura clara. - **Instalación**: ejecutar pasos de setup y validaciones básicas de herramientas desde flujos guiados por CLI. - **Reparación de entorno**: diagnosticar fallos típicos (dependencias, rutas, comandos) y aplicar correcciones verificables. - **Automatización web**: abrir sitios, extraer datos concretos y guardar resultados sin depender de interacción manual constante. > **Si esto te ahorra tiempo real, estás en el camino correcto.** ## Archived - [legacy/](D:/IA/CODE/claudebot/legacy): previous prototypes not used by the current flow. **Status: archived** ## Minimal Structure ```text claudebot/ |-- orchestrator/ active project |-- tools/ external utilities and helper repos |-- legacy/ archived prototypes |-- self_model/ shared routing self-model |-- episodic_memory/ shared execution episodes |-- world_model/ shared desktop world state `-- .claude/ local config ``` ## Natural Language Automation The main orchestrator now auto-routes plain language desktop and browser tasks. You do not need to remember prefixes like `browser:` or `windows:`. Examples: ```bash cd D:/IA/CODE/claudebot/orchestrator run_orchestrator.bat "Abre Chrome y ve a https://example.com" run_orchestrator.bat "Abre Notepad y escribe hola mundo" run_orchestrator.bat "Abre https://example.com y guarda un resumen en tasks/output/resumen.txt" ``` Behind the scenes: - browser tasks go to `worker-core:browser` - Windows desktop tasks go to `worker-core:windows` - multi-step automation flows go to `worker-core:orchestrator` The skills taxonomy used by these wrappers is documented in [docs/SKILLS_CATALOG.md](D:/IA/CODE/claudebot/docs/SKILLS_CATALOG.md). Recommended high-level entrypoint: ```bash python run_agent.py --agent claude_code --tool auto "Abre https://example.com y extrae el titulo" python run_agent.py --agent codex_cli --tool browser "Abre https://example.com y extrae el titulo" python run_agent.py --agent minimax_cli --tool worker "Abre https://example.com y guarda un resumen en tasks/output/resumen.txt" ``` You can still call the low-level wrappers directly if you want explicit control: ```bash python run_browser.py "Abre https://example.com y extrae el titulo" python run_windows.py "Abre Notepad y escribe hola mundo" python run_worker.py "Abre https://example.com y guarda un resumen en tasks/output/resumen.txt" ``` ## Shared Skill Bridge The workspace now includes a local bridge that can discover and reuse skills across: - `gemini_skills/` for compatible executable skills, - `skills/` and `skills/.codex/skills/` for `SKILL.md`-based skills, - `legacy/local-agent/SKILLS/`, - packaged skills under `tools/browser-use/skills/` and related tooling. Examples: ```bash python skills_bridge.py list gemini --limit 10 python skills_bridge.py suggest "Necesito automatizar browser y extraer datos" python skills_bridge.py show gemini.mouse python skills_bridge.py run gemini.mouse -- --request-json "{\"x\":0.5,\"y\":0.8,\"coordinate_space\":\"normalized\",\"action\":\"move\"}" python gemini_bridge.py skills gemini --limit 10 python gemini_bridge.py skill-info gemini.chatgpt --content python gemini_bridge.py skill-run gemini.vision_world ``` Operational rule: - `instruction` skills are discoverable and readable through the bridge. - `executable` skills can be launched through the bridge. - The orchestrator now injects relevant skill matches into model context alongside self-model, episodic memory, and world model. For direct coordinate-driven desktop control with calibration and verification: ```bash python run_mouse.py 960 540 --action click --label "boton enviar" --verify cursor python run_mouse.py 640 360 --action click --coordinate-space image --image-path screenshot.png --verify screen_change python run_mouse.py --request-json "{\"x\":0.5,\"y\":0.8,\"coordinate_space\":\"normalized\",\"action\":\"move\"}" ``` `run_mouse.py` stores a learned calibration profile per app and resolution, retries around the predicted point, and records successes/failures into the shared self-model and episodic memory. Status: **deprecated** as default strategy; keep only as explicit fallback for cases where CLI/browser/OS routes are not viable. ## Telegram Voice Notes `telegram_bridge.py` now accepts Telegram `voice` and `audio` messages. The bridge downloads the note, transcribes it locally with Whisper, and then sends the transcript through the same Gemini CLI flow used by plain text messages. Recommended local runtime: - `faster-whisper` on CTranslate2. - Default bridge config is `tiny` + `cpu` + `int8` to keep GPU impact close to zero while still supporting Spanish and other multilingual voice notes. - If you want a newer and more accurate model later, set `TELEGRAM_WHISPER_MODEL=turbo`, but expect much higher memory usage. Install: ```bash pip install faster-whisper ``` Optional `.env` overrides: ```bash TELEGRAM_WHISPER_MODEL=tiny TELEGRAM_WHISPER_DEVICE=cpu TELEGRAM_WHISPER_COMPUTE_TYPE=int8 TELEGRAM_WHISPER_LANGUAGE=es TELEGRAM_WHISPER_BEAM_SIZE=1 TELEGRAM_WHISPER_DOWNLOAD_ROOT=D:/IA/CODE/claudebot/models_stt/whisper TELEGRAM_WHISPER_VAD_FILTER=true ``` ## Capacidades avanzadas (secundario) Estos componentes son útiles, pero no son el núcleo del valor diario: - `self_model/` para conocimiento acumulado de routing. - `episodic_memory/` para reusar aprendizajes de ejecuciones previas. - `world_model/` para estado operativo compartido entre agentes. Si tu caso de uso principal es productividad cotidiana, empieza por el flujo CLI-first y añade estas capas cuando ya tengas una rutina estable. ## Shared Self-Model The repo now includes a live `self_model/` directory at the root. This is the explicit model of how the system understands itself. It is shared across: - Claude-style orchestrator runs - Gemini bridge runs - Codex or other CLI agents using the root wrappers Examples: ```bash python self_model_cli.py summary python self_model_cli.py plan "Fix this broken project and validate the output" python gemini_bridge.py auto "Abre Chrome y ve a https://example.com" python run_browser.py --agent codex_cli "Abre https://example.com y extrae el titulo" ``` The key point is that routing decisions and outcomes now update the same structured state instead of living only in ad-hoc logs. ## Episodic Memory The repo now also includes a separate `episodic_memory/` layer. This is not the self-model. - `self_model/` answers "what am I generally good or bad at?" - `episodic_memory/` answers "what happened last time in a similar task, app, page, or failure?" Each episode can store: - attempted task - executed steps - app, web, and screen context - exact failure - working fix or successful route Examples: ```bash python episodic_memory_cli.py summary python episodic_memory_cli.py find "Abre Chrome y ve a https://example.com y revisa el login" --task-type browser_automation ``` Claude-style orchestrator runs, Gemini bridge runs, and the root wrappers all write into the same episodic layer. ## World Model The repo now also includes a separate `world_model/` layer. This is the live desktop/browser/filesystem state that agents can consult before acting. It tracks things like: - open apps - active window - files created or touched - downloads in progress - browser tabs associated with tasks - pending objectives Examples: ```bash python world_model_cli.py summary --refresh python world_model_cli.py focus "Abre Chrome y ve a https://example.com y guarda un resumen" ``` The main orchestrator, Gemini bridge, and root wrappers all update the same shared world state.

README des Repositorys wird geladen…

Skills in diesem Repository

Softwareentwickler Musikdirektoren und Komponisten Spezialeffektkünstler und Animatoren Artdirektoren

12 Berufskategorien · 100% klassifiziert

Mit einem KI-Assistenten installieren

Diesen Prompt kopieren und in den verwendeten KI-Assistenten einfügen.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/futurediffusion/claudebot.