HKUDS/CLI-Anything

CLI-Anything is a framework for generating and managing agent-native command-line interfaces that wrap existing desktop applications, web services, and development tools. It provides a standardized structure for exposing complex software capabilities—such as 3D modeling, video editing, and database management—through consistent, machine-readable JSON APIs and interactive REPLs. The system includes a central hub for discovering these harnesses, plugins for AI coding agents to build them, and a registry of skill definitions to facilitate automated tool selection.

An evaluation harness has been added for the Audacity CLI to validate core functionality. This includes specific tasks for verifying the effects registry (checking for normalize and fade\_in), exporting projects to WAV files, performing project roundtrips (create, save, open, and validate metadata), and managing the track and clip flow (adding tracks, clips, and splitting clips).

_audacity/agent-harness/cli\anything/audacity/eval · high confidence

A new command-line interface (CLI) harness for CC Switch has been added, allowing users and AI agents to manage AI coding tool configurations (Claude Code, Codex, Gemini CLI, OpenCode, OpenClaw, Hermes) directly from the terminal. The CLI reads and writes to the live CC Switch SQLite database and config files, supporting operations such as listing and switching providers, managing MCP servers and skills, viewing usage stats, and adjusting proxy and session settings. It includes sensitive data masking for API keys and tokens, JSON output mode for programmatic consumption, and comprehensive unit and end-to-end tests.

cc-switch · high confidence

Introduced the 'CLI-Anything' skill for Hermes Agent, enabling the agent to build, refine, test, and validate CLI harnesses for GUI applications or source repositories. The skill includes a detailed instruction set (SKILL.md) covering build, refine, test, and validate modes, along with agent configuration (agents/hermes.yaml) and installation scripts for both Unix (install.sh) and Windows (install.ps1) environments.

hermes-skill · high confidence

The \comfyui/utils/comfyui\_backend.py\ module has been added to provide a dedicated HTTP client for interacting with the ComfyUI server. This utility wraps standard REST API calls (GET, POST, DELETE) and raw byte retrieval, handling connection to the local server (defaulting to \http://localhost:8188\) and raising descriptive errors for timeouts or network failures.

_comfyui/agent-harness/cli\anything/comfyui/utils · high confidence

A new empty Python package \cli\_anything\ has been added under the \comfyui/agent-harness\ directory, establishing the initial structure for a CLI-based agent harness integration with ComfyUI. This change introduces the module namespace but does not yet include any functional code or user-facing features.

_comfyui/agent-harness/cli\anything · medium confidence

Introduces a new skill that enables Reasonix to build, refine, test, and validate CLI harnesses for GUI applications using the CLI-Anything methodology. The adapter includes installation scripts for macOS/Linux and Windows, a skill definition with specific tool bindings (such as bash, write\_file, and optional CodeGraph integration), and guidance on step budgets to ensure complex builds complete successfully.

reasonix-skill · high confidence

Introduces a new Python-based CLI harness (\cli-anything-tigris\) that wraps the official Tigris CLI to provide AI agents with structured access to Tigris object storage. This tool exposes Tigris-specific primitives—including bucket management, object operations, presigned URLs, point-in-time snapshots, IAM policies, and scoped access keys—through a unified interface with consistent \--json\ output for machine parsing. It requires the \tigris\ CLI binary on the PATH and supports both interactive REPL and direct command-line usage, enabling agents to perform server-side copies, manage credentials, and handle destructive actions safely with explicit confirmation flags.

tigris · high confidence

Introduces a new CLI harness for Zoom, enabling users to manage meetings, participants, and recordings via the command line. The tool supports OAuth2 authentication, full meeting lifecycle operations (create, list, update, delete, join, start), participant management, and recording retrieval. It includes an interactive REPL mode and JSON output for agent integration.

_zoom/agent-harness/cli\anything/zoom · high confidence

A new template file (SKILL.md.template) has been added to the cli-anything-plugin to generate standardized documentation for CLI skills. This template structures information including installation prerequisites, basic and REPL usage commands, command group tables, examples, state management details, and specific instructions for AI agents (such as using the --json flag). It uses Jinja2 variables to dynamically populate content like skill names, descriptions, and version numbers, facilitating consistent discovery and usage of CLI tools by both humans and AI agents.

cli-anything-plugin/templates · high confidence

The new setup-qodercli.sh script in the qoder-plugin directory allows users to automatically register the cli-anything plugin with their Qodercli configuration. The script validates the plugin directory, ensures the config file is valid JSON, and safely updates or creates the .qoder.json file to include the plugin path, providing a backup of the original config if it already exists.

qoder-plugin · high confidence

The Blender CLI now supports capturing and inspecting real-time preview bundles via the new \preview\ command group, including a \live\ subcommand that manages persistent preview sessions with background polling and optional viewer integration. Additionally, one-shot commands now automatically save the project state upon exit unless the \--dry-run\ flag is specified, ensuring changes are persisted by default.

_blender/agent-harness/cli\anything/blender · high confidence

The Blender harness now supports live preview sessions, introducing a new preview module that handles bundle generation, live polling, and trajectory recording for real-time scene review. To ensure data integrity during concurrent operations, session state is now saved using atomic file locking, preventing race conditions when multiple processes access the session file. Additionally, render output paths are forced to be absolute to prevent failures when the Blender background process runs with a different working directory.

_blender/agent-harness/cli\anything/blender/core · high confidence

The cli-anything plugin now generates SKILL.md files (Phase 6.5) to make CLIs discoverable by AI agents, with auto-detection of the skill path in the REPL banner. It introduces a preview bundle system (preview\_bundle.py) for caching and managing preview trajectories. Session management now uses atomic file locking to prevent race conditions. The license has been updated from MIT to Apache 2.0. Documentation and verification scripts have been updated to reflect these changes, including the addition of the /cli-anything:list command and the removal of stale build.md references.

cli-anything-plugin · high confidence

The LibreOffice CLI harness now supports importing existing ODF and Office files (such as .docx) into its JSON project model for editing, in addition to creating new documents from scratch. To ensure safe handling of potentially untrusted XML content during these operations, the harness now relies on the defusedxml library for XML parsing. The package version has been updated to 1.0.1 to reflect these changes.

libreoffice/agent-harness · high confidence

Users can now import existing Office and ODF files (such as DOCX, XLSX, PPTX, and ODT) into the CLI project model via a new importer module that handles both native ODF parsing and conversion via LibreOffice headless. To ensure data integrity, session saves now use atomic file locking to prevent race conditions, and JSON keys are sorted for deterministic output. Additionally, opening project files now provides clearer error messages if the JSON is invalid or corrupted.

_libreoffice/agent-harness/cli\anything/libreoffice/core · high confidence

The repository is initialized with core project metadata and contributor guidelines. A CITATION.cff file is added to provide standardized academic citation details for the CLI-Anything technical report. Comprehensive CONTRIBUTING.md documentation is introduced, outlining the workflow for adding new CLI harnesses (both in-repo and standalone), submitting features, and reporting bugs, including specific requirements for registry.json entries and test coverage. The project is licensed under Apache 2.0 via a new LICENSE file. Additionally, localized README files for Chinese (README\_CN.md), German (README\_DE.md), and Japanese (README\_JA.md) are added to support international users, and a SECURITY.md policy is established to define threat models and reporting procedures for vulnerabilities. A matrix\_registry.json file is also included to support multi-CLI workflow definitions.

(repo-wide) · high confidence

The Inkscape CLI harness now supports automatic saving of project changes after each command in one-shot mode via the new -s/--save flag, and includes a --dry-run option to preview mutations without persisting changes. It also introduces optional text box width, height, and line-height parameters for the text-add command to enable text wrapping, and improves REPL input parsing by using shlex.split for more robust argument handling.

_inkscape/agent-harness/cli\anything/inkscape · high confidence

The \cli-anything-3mf\ tool is now available for inspecting, modifying, and repairing 3MF mesh files. It enables feature-aware geometry editing, such as detecting cylindrical holes and resizing them via CLI commands, while preserving slicer metadata (BambuStudio, PrusaSlicer) and triangle attributes. The tool also includes mesh repair capabilities to fix degenerate faces and duplicate vertices.

3MF · high confidence

Adds a new command-line interface for the AnyGen cloud service, enabling users to generate slides, documents, diagrams, and other content via natural language prompts. The CLI wraps the AnyGen OpenAPI, providing commands to create tasks, poll for completion, and download generated files, along with support for file uploads and multi-turn requirement analysis. It includes a stateful session with undo/redo history, local task persistence, and dual output modes (human-readable and machine-readable JSON) for agent consumption.

anygen · high confidence

Added a new Python package, cli-anything-comfyui, which provides a command-line interface for managing AI image generation workflows via the ComfyUI REST API. The package requires ComfyUI to be running locally at http://localhost:8188 and exposes a console script named cli-anything-comfyui. It depends on click and requests, supports Python 3.10+, and includes skill documentation files within the package data.

comfyui/agent-harness · high confidence

Users can now automate repetitive data entry for Japan Post Web Yu-pri shipping labels via a new Python CLI (\cli-anything-web-yu-pri\). The tool drives the real browser UI through Playwright using a persistent local profile, allowing users to log in manually once and then run structured commands for form entry. It accepts item lists in JSON, CSV, or TSV formats and supports commands to validate plans, inspect page status, capture screenshots, and fill contents rows. For safety, the tool does not store passwords and stops after filling contents and totals, never clicking the final shipment confirmation button.

web-yu-pri · high confidence

Added a new command-line interface for Krita that allows users to manage projects, layers, filters, and exports without the GUI. The harness uses a JSON-based project state to track document structure and leverages Krita's headless export capabilities to render final images, supporting formats like PNG, JPEG, TIFF, PSD, and more. It includes features for undo/redo history, auto-save for one-shot commands, and an interactive REPL for complex workflows.

krita · high confidence

This change adds a new command-line interface for Mermaid Live Editor, enabling users to create, edit, and render Mermaid diagrams directly from the terminal. The harness supports stateful project management via JSON files, allowing users to create new projects with sample presets, update diagram source text, and inspect current state. It provides export capabilities to render diagrams as SVG or PNG artifacts using the official Mermaid renderer service, and generates shareable URLs for the Mermaid Live Editor. The tool includes an interactive REPL mode for iterative workflow, supports undo/redo for source changes, and offers machine-readable JSON output for automation. This is a new capability rather than a modification of existing behavior.

mermaid · high confidence

Adds a new command-line interface for the Openscreen screen recording editor, allowing users to create, open, and save projects, and to edit timelines by adding zoom, speed, trim, and annotation regions, as well as setting crop, background, and export options. The harness uses an ffmpeg backend to render final videos and supports a preview bundle feature for generating low-resolution review clips and sampled frames. It includes a stateful session with undo/redo, auto-save, and dry-run modes, and is accompanied by 101 unit and end-to-end tests.

openscreen · high confidence

Added a new command-line interface for the Teltonika Remote Management System (RMS) API, enabling users to manage devices, companies, users, alerts, configurations, and more directly from the terminal. The tool supports interactive REPL mode, JSON output, and persistent session state, covering resources such as device listing, alert management, credit transfers, and file uploads.

rms · high confidence

Adds the \cli-anything-zotero\ package, providing an agent-native command-line interface for Zotero desktop. The harness composes Zotero's existing local surfaces—SQLite for offline inventory, the connector API for GUI-aware writes (importing literature, attaching PDFs, adding notes), and the Local API for live search, citation rendering, and bibliography export. It includes core modules for catalog management, item analysis (with optional OpenAI integration), and session state persistence, along with documentation and agent skill constraints.

zotero · high confidence

Adds a new Codex skill that enables building, refining, testing, validating, and listing CLI-Anything harnesses for GUI applications or source repositories. The skill includes a configuration file for the OpenAI agent interface and platform-specific install scripts (PowerShell for Windows, Bash for Unix) that vendor the full CLI-Anything methodology resources (HARNESS.md, command specs, guides, scripts, and preview protocol) into the installed skill directory. The installer uses a staging directory to ensure atomic installation and refuses to overwrite existing skills, preventing data loss. Tests verify that all resources are correctly copied, Python scripts compile, stale staging directories are handled, and detached installs fail gracefully when canonical resources are missing.

codex-skill · high confidence

Adds a new Cursor plugin that provides slash commands to build, refine, test, validate, and list CLI-Anything harnesses. The plugin includes an installer (install.sh/install.ps1) that vendors the CLI-Anything methodology and helper scripts into the local Cursor plugins directory, along with uninstall scripts and a discovery pointer to ensure Cursor tools can resolve the correct paths. This generator capability is separate from the consumer Hub path for finding and installing published CLIs.

cursor-plugin · high confidence

The .pi-extension/cli-anything directory now contains the full CLI-Anything extension for the Pi Coding Agent, including the main entry point (index.ts), a global installation script (install.sh), documentation (README.md), and unit tests (tests/test\_extension.test.ts). The extension registers five slash commands (/cli-anything, /cli-anything:refine, /cli-anything:test, /cli-anything:validate, /cli-anything:list) that inject HARNESS.md methodology and command specifications into the agent session, handle path remapping between containerized and local environments, and provide usage warnings when arguments are missing.

.pi-extension · high confidence

The dify-workflow area now includes a new \cli-anything-dify-workflow\ harness that wraps the upstream \dify-workflow\ CLI. This wrapper exposes the upstream command surface (guide, list-node-types, create, inspect, validate, edit, config, export, import, diff, layout) through the CLI-Anything ecosystem, enabling AI agents to discover and use the tool via CLI-Hub. It provides a unified REPL skin for interactive use, forwards commands to the installed upstream CLI (with a fallback to the Python module if the binary is not on PATH), and includes tests to verify wrapper discovery and forwarding behavior.

dify-workflow · high confidence

Adds a new CLI harness (\cli-anything-calibre\) that wraps the real Calibre binaries (\calibredb\, \ebook-convert\, \ebook-meta\) to provide a structured interface for AI agents and scripts. This tool enables library management, metadata editing, and format conversion without reimplementing Calibre's core logic. The package includes a Python setup configuration, session-based state management, and comprehensive documentation outlining supported commands and workflows.

calibre/agent-harness · high confidence

Added a new command-line interface for interacting with ChromaDB vector databases, enabling users to manage collections and documents and perform semantic search directly from the terminal. The tool provides an interactive REPL mode and supports direct subcommands for server health checks, collection CRUD operations, document management, and semantic queries. It communicates with the ChromaDB HTTP API v2, supports configurable server hosts, and offers both human-readable formatted output and machine-readable JSON output for automation.

chromadb, pm2, seaclip · high confidence

A new CLI harness (\cli-anything-cloudanalyzer\) wraps the CloudAnalyzer Python package to provide a structured, agent-friendly interface for point cloud quality assurance. It exposes 27 commands across eight groups (evaluation, trajectory, checks, baselines, processing, inspection, info, and session) via a unified Click interface, ensuring every command supports \--json\ output for automated consumption. The harness includes an interactive REPL, project and session state management with undo/redo capabilities, and a \SKILL.md\ file for agent auto-discovery. It connects directly to the CloudAnalyzer backend via Python imports rather than subprocesses, and ships with unit tests for core logic and E2E tests for full command execution.

cloudanalyzer · high confidence

Added a new agent-friendly command-line interface for CloudCompare, wrapping its native \-SILENT\ mode to enable programmatic 3D point cloud and mesh processing. The harness provides a structured CLI with command groups for project management, session control, point cloud operations (such as subsampling, filtering, and analysis), mesh operations, distance computation (cloud-to-cloud and cloud-to-mesh), transformations (including ICP registration), and format export. It supports an interactive REPL, JSON output for machine consumption, and handles CloudCompare detection across native, Flatpak, and Snap installations. The release includes comprehensive unit and end-to-end test suites to validate core functionality and real-world workflows.

cloudcompare · high confidence

Adds a new CLI tool (\cli-anything-comfyui\) that wraps the ComfyUI REST API, enabling users to manage AI image generation workflows directly from the terminal. Users can now list and validate workflow files, queue prompts for generation, monitor queue status, discover available models (checkpoints, LoRAs, etc.), and download generated images. The tool supports both human-readable output and machine-parseable JSON mode (\--json\) for automation, and includes an interactive REPL for exploratory use.

_comfyui/agent-harness/cli\anything/comfyui · high confidence

This change adds the core Python modules for the new ComfyUI agent harness, enabling programmatic interaction with a ComfyUI server. The \models.py\ module allows users to discover available checkpoints, LoRAs, VAEs, and ControlNet models, as well as inspect node schemas via the \/object\_info\ endpoint. The \queue.py\ module provides functions to submit workflows, monitor queue status, retrieve generation history, and interrupt running tasks. Additionally, \workflows.py\ handles loading, saving, listing, and validating local workflow JSON files, while \images.py\ manages downloading generated images from the server to local storage.

_comfyui/agent-harness/cli\anything/comfyui/core · high confidence

This change adds the core Python module for a new Draw.io CLI harness, enabling agents to programmatically create and edit diagrams. The implementation provides a stateful session management system with undo/redo capabilities and atomic file locking for session persistence. It exposes operations for managing project pages (add, remove, rename), manipulating shapes (add, remove, move, resize, style, label), and connecting shapes with edges (add, remove, style, label). Additionally, it supports exporting diagrams to multiple formats (PNG, PDF, SVG, VSDX, XML) via the draw.io CLI, with a fallback mechanism to save the raw .drawio file if the CLI is unavailable.

_drawio/agent-harness, drawio/agent-harness/cli\anything/drawio/core · high confidence

Added a new command-line interface for Draw.io that allows users and AI agents to create, edit, and export diagrams programmatically. The tool supports one-shot commands for adding shapes, connecting elements, and exporting to PNG, PDF, or SVG, as well as an interactive REPL mode. It includes features like JSON output for agent integration, session management, and auto-save functionality, enabling seamless diagram generation from the command line.

_drawio/agent-harness/cli\anything/drawio · high confidence

Added a new CLI harness for creating and exporting Draw.io diagrams, including utilities to parse and manipulate the underlying XML structure and a backend module to invoke the draw.io desktop CLI for exporting diagrams to PNG, PDF, SVG, or VSDX. To address security concerns regarding untrusted XML input, the XML parsing logic now uses the \defusedxml\ library instead of the standard library, preventing potential XML external entity (XXE) attacks.

_drawio/agent-harness/cli\anything/drawio/utils · high confidence

Adds a new CLI harness (\cli-anything-exa\) that wraps the Exa API to provide neural web search and full-text content extraction for AI agents. The tool supports multiple search modes (auto, fast, deep, deep-reasoning), domain filtering, and content retrieval (highlights, text, summary) with both human-readable and JSON output modes. It includes an interactive REPL for agent workflows, a connectivity check command, and a structured skill definition for agent discovery. The implementation relies on the \exa-py\ SDK and includes comprehensive unit and end-to-end tests.

exa · high confidence

The firefly-iii agent-harness now provides a stateless command-line interface (CLI) built on the CLI-Anything specification, replacing the previous long-running MCP mode to avoid Node residual process issues. This harness installs as the \cli-anything-firefly-iii\ Python package and exposes comprehensive command groups for managing accounts, transactions, budgets, bills, categories, tags, piggy banks, currencies, rules, and recurrences, along with insights, search, and data export capabilities. Users can interact via REPL or subcommand modes, utilizing preset filtering (\default\, \full\, \budget\, etc.) and structured JSON output, while configuration is handled through environment variables or command-line arguments for the Firefly III base URL and Personal Access Token.

firefly-iii · high confidence

Adds a new agent-native CLI harness for the Godot Engine (4.x) that enables programmatic creation and inspection of projects, scenes, and resources, as well as GDScript execution and platform exports. The tool wraps Godot's headless CLI flags via subprocess calls, providing structured JSON output for AI agents and a unified REPL interface. It includes core modules for project metadata, scene file I/O, export preset handling, and script validation, along with comprehensive unit and end-to-end tests covering the full demo-game pipeline.

godot · high confidence

The \joplin/agent-harness\ directory now contains the \cli-anything-joplin\ harness, a stateful CLI tool that wraps the real \joplin\ terminal binary. It provides a unified JSON command envelope for managing notebooks, notes, to-dos, tags, attachments, sync, and configuration, while maintaining an in-memory session with undo/redo and auto-save capabilities. The harness includes a full test suite (107 unit tests and 134 total tests) and handles backend-specific edge cases, such as probing for optional flags like \--permanent\ and \--exit-early\ to prevent silent failures on older Joplin CLI versions, and falling back to \package.json\ metadata when the native \version\ command fails in certain npm global layouts.

joplin · high confidence

Adds a new Python-based CLI harness (cli-anything-jumpserver v0.1.0) for managing JumpServer bastion host operations. This tool provides a unified command-line interface for asset management (hosts, devices, databases, nodes), user and permission management, account and credential handling, session monitoring and replay, audit log viewing, and system health checks. It supports interactive REPL mode, multiple output formats (table, JSON, YAML), and includes a dry-run capability for safe previewing of mutation commands.

jumpserver · high confidence

Adds a new LLDB agent harness (\cli-anything-lldb\) that wraps the LLDB Python API for structured debugging workflows. The CLI supports persistent, stateful sessions across separate command invocations via a localhost JSON socket daemon with owner-scoped state files, and includes a formal stdio Debug Adapter Protocol (DAP) server for AI and editor integration. Key capabilities include target creation, process launch/attach (including PID-only attach), breakpoint management with honest resolution reporting, thread/frame inspection, expression evaluation, and memory operations. The DAP server features non-blocking \continue\ for long-running GUI targets, async pause support, guarded breakpoint mutation during active execution, and structured stop-rule profiles to auto-continue internal traps (e.g., NVIDIA JIT, Windows debug breaks). The harness ships with comprehensive unit and end-to-end tests covering DAP lifecycle, session persistence, and debugger operations.

lldb · high confidence

MacroCLI is a new layered CLI that converts GUI workflows into parameterized, agent-callable macros. Agents invoke \cli-anything-macrocli macro run \<name\>\ to execute steps that are routed through a priority-based engine to one of seven backends: native API (subprocess), GUI macro (precompiled coordinate replay), visual anchor (template matching), file transform (XML/JSON/text editing), GUI agent (vision-model-driven automation via OpenAI-compatible APIs), semantic UI (accessibility APIs/xdotool), and recovery (retry/fallback). Macros are defined in YAML with parameter validation, pre/postconditions, and telemetry, and the system handles execution, state, and structured result output without the agent touching the GUI directly.

macrocli · high confidence

A new Python CLI tool, \cli-anything-mailchimp\, is added to the \mailchimp/agent-harness\ directory, exposing the full Mailchimp Marketing API v3.0 as 303 typed Click commands. The tool is built on the CLI-Anything framework and uses a code generator (\\_codegen/generate.py\) to create command modules from the official Swagger 2.0 spec. It supports all major resource groups including lists, campaigns, automations, and ecommerce, with features like JSON output, REPL mode, and environment-variable-only authentication via \MAILCHIMP\_API\_KEY\.

mailchimp · high confidence

Adds a new \cli-anything-minimax\ command-line tool that provides chat completions (defaulting to the MiniMax-M3 model) and text-to-speech synthesis via the MiniMax API. The harness includes a Click-based CLI with subcommands for chat, streaming, TTS, session management, and configuration, along with a local session history store and support for regional endpoints (global and China). TTS output is fully configurable, exposing voice selection and audio parameters such as speed, volume, pitch, sample rate, bitrate, format, and channel count. The package ships with unit and end-to-end tests, including subprocess smoke tests and mocked API validation, and installs as a standalone console script.

minimax · high confidence

The mubu location now ships a packaged agent harness (cli-anything-mubu v0.1.1) that provides a Click-based CLI and an interactive REPL for inspecting and mutating the local Mubu desktop workspace. Users can discover documents and folders, inspect live nodes, and perform atomic mutations (create-child, delete-node, update-text) via the live Mubu API, with mutations defaulting to dry-run and requiring explicit --execute. The harness persists session state (current document, current node, command history) in \~/.config/mubu-cli (falling back to the legacy cli-anything-mubu path) and supports daily-note helpers via an explicit folder reference or the MUBU\_DAILY\_FOLDER environment variable. A skill generator can regenerate the packaged SKILL.md, and the harness includes a comprehensive test suite covering unit, integration, and end-to-end workflows.

mubu · high confidence

Adds a new \cli-anything-musescore\ tool that wraps the MuseScore 4 backend (\mscore\) to enable music notation tasks from the command line. Users can now transpose scores by key, interval, or diatonically; export scores to multiple formats including PDF, PNG, SVG, MP3, FLAC, WAV, MIDI, MusicXML, and Braille; extract individual parts from multi-instrument scores; manage instruments (list, add, remove, reorder); and analyze scores via metadata, statistics, and diffing. The tool supports both an interactive REPL with undo/redo session management and one-shot JSON-output commands, and includes a comprehensive test suite verifying core logic and end-to-end workflows.

musescore · high confidence

Users can now interact with Novita AI models (such as DeepSeek and GLM) via the new \cli-anything-novita\ command-line tool. This harness provides one-shot chat and streaming commands, an interactive REPL with persistent session history, and machine-readable JSON output for agent consumption. It supports configuration via environment variables or a local config file, includes connectivity testing, and lists available models.

novita · high confidence

Adds a new Windows-first CLI harness (\cli-anything-nsight-graphics\) that wraps official NVIDIA Nsight Graphics tools to orchestrate installation probing, detached launch, PID attach, Graphics Capture, OpenGL Frame Debugger capture, GPU Trace capture with auto-export and summary generation, replay analysis of existing \.ngfx-capture\ files, and Generate C++ Capture. The harness supports both legacy unified (\ngfx.exe\) and modern split (\ngfx-capture\/\ngfx-replay\) tool layouts, automatically selecting the newest complete GPU Trace export when multiple are present and providing structured JSON output for programmatic workflows.

nsight-graphics · high confidence

Adds a new command-line interface for interacting with the Obsidian Local REST API, enabling users to manage vault files (list, read, create, update, delete, append), perform structured searches via Dataview DQL or JsonLogic, and execute Obsidian commands. The tool includes an interactive REPL mode, supports JSON output for agent consumption, and handles authentication via Bearer tokens. This harness allows AI agents and power users to control Obsidian without the GUI.

obsidian · high confidence

Adds a new command-line interface harness for Ollama, enabling users to manage local models (list, pull, delete, copy), generate text, run multi-turn chats, and create embeddings via the Ollama REST API. The tool supports both one-shot commands and an interactive REPL, with output available in human-readable tables or machine-readable JSON for agent consumption. It includes a full test suite covering unit tests for CLI parsing and backend URL construction, as well as end-to-end tests against a live Ollama server.

ollama · high confidence

Adds a new \cli-anything-openrefine\ CLI harness that exposes OpenRefine's local HTTP API for agent-driven data wrangling, allowing users to import projects, inspect rows, apply operation histories, export cleaned data, and manage session state via a Click-based interface. The harness supports an interactive REPL and, crucially, enables piped stdin workflows for CI and automation by automatically switching to ASCII-only output when redirected, ensuring reliable execution in non-interactive environments like Windows legacy encodings or CI runners.

openrefine · high confidence

Added a new command-line interface (cli-anything-renderdoc) in renderdoc/agent-harness that wraps the RenderDoc Python API to enable scriptable, headless inspection of GPU frame captures (.rdc files) without the GUI. The harness provides subcommands for capture metadata and conversion, action/draw-call enumeration and filtering, texture listing and pixel picking, full pipeline state and shader disassembly inspection, resource and buffer data reading, mesh input/output decoding, and GPU performance counter fetching. It supports JSON output for automation, lazy-loading of the RenderDoc module so help works without a full install, and includes unit and end-to-end tests alongside documentation.

renderdoc · high confidence

Added a new \safari/agent-harness\ that wraps the \safari-mcp\ Node.js MCP server in a Python Click CLI (\cli-anything-safari\), enabling Safari browser automation on macOS for non-MCP agent frameworks, bash scripts, and CI pipelines. The harness exposes all 84 tools from \safari-mcp\ via a schema-driven CLI where commands are auto-generated from a bundled \tools.json\ registry, ensuring feature parity and argument accuracy. It includes URL validation to block dangerous schemes, a \--json\ output flag, an interactive REPL, and a comprehensive test suite (unit, security, parity, and gated E2E) to verify correctness and security.

safari · high confidence

Adds a new Python-based CLI tool (\cli-anything-siyuan\) that allows users to manage SiYuan notebooks, documents, and blocks directly from the terminal. The harness connects to a running SiYuan instance via its HTTP API (default port 6806) to perform operations such as creating, moving, and deleting documents, inserting and updating blocks, managing tags, and executing SQL queries. It supports piping multi-line content for block operations and includes documentation on usage patterns and API capabilities.

siyuan/agent-harness · high confidence

Adds a new CLI tool (\cli-anything-siyuan\) that connects to a running SiYuan instance (v3.x) via its HTTP API, enabling users to manage notebooks, documents, and blocks, perform full-text search, execute SQL queries, and export content as Markdown. The tool supports both one-shot commands and an interactive REPL mode, with features such as stdin piping for block content (handling CJK characters via UTF-8-sig decoding), JSON output formatting, and persistent session state across REPL invocations.

_siyuan/agent-harness/cli\anything/siyuan · high confidence

Adds a new CLI harness for SiYuan, including a backend module that locates the SiYuan data directory, verifies connectivity to a running instance, and retrieves the API token from platform-specific configuration files. It also introduces a unified REPL skin component that provides consistent terminal branding, colored prompts, and status messages across all CLI harnesses.

_siyuan/agent-harness/cli\_anything/siyuan/core, siyuan/agent-harness/cli\anything/siyuan/utils · high confidence

Users can now control the real Steam version of Slay the Spire II via a command-line interface (\cli-anything-sts2\) rather than a headless simulator. This requires installing a custom \.NET 9\ bridge mod (\STS2\_Bridge\) that exposes the game's internal state and actions through a local HTTP API at \localhost:15526\. The CLI reads normalized JSON state and sends commands for menu navigation, combat actions (playing cards, using potions), map selection, and reward claiming. The bridge supports both single-player and multiplayer modes, including specific multiplayer actions like ending and undoing turns.

_slay\_the\_spire\ii · high confidence

Adds a new CLI-Anything harness for Uni-Mol Tools that provides an interactive REPL and JSON API for molecular property prediction. This location introduces the core orchestration layer, including project management (create, load, save), training workflows, and a comprehensive model management suite that supports ranking, comparison, storage analysis, and cleanup of trained models. The harness also includes a demo script to exercise all five supported task types (binary classification, regression, multiclass, multilabel classification, and multilabel regression) and a full test suite to verify functionality.

_unimol\tools · high confidence

This change adds a new \cli-anything-unrealinsights\ harness that wraps the real Unreal Engine \UnrealInsights.exe\ and target executables to provide agent-friendly, command-line access to performance profiling workflows. Users can now discover backend binaries, browse the local Trace Store, launch traced targets for file-mode capture, and control continuous capture sessions (start, status, snapshot, stop). The harness supports headless analysis of \.utrace\ files via exporters for threads, timers, timing events, counters, and more, including legacy UE 5.3 compatibility handling for export paths. It also enables live trace control (status, bookmarks, screenshots, stop) through an external backend template, GUI co-pilot commands, and automated summary analysis of timing/counter hotspots and anomalies. Comprehensive unit and E2E tests are included to validate these capabilities.

unrealinsights · high confidence

This change adds a new CLI harness for VideoCaptioner, wrapping the production \videocaptioner\ backend via subprocess calls. It provides a unified interface for transcribing audio/video to subtitles (supporting bijian, jianying, whisper-api, and whisper-cpp engines), optimizing and translating subtitles (with LLM, Bing, and Google translators), and synthesizing videos with soft or hard subtitles. The harness includes a full pipeline command, a review mode to check subtitle/script consistency before hard-burning, and JSON output for agent consumption. It also ships with unit and end-to-end tests to verify argument construction and backend integration.

videocaptioner · high confidence

Adds a new \cli-anything-wavetone\ package that provides an agent-facing CLI interface for WaveTone 2.61. The harness creates and manages JSON project manifests to track audio files, analysis settings, labels, and tempo metadata, while probing audio metadata via Python's \wave\ module or \ffprobe\. It launches the real \wavetone.exe\ backend for analysis and export, which remain GUI-driven, and includes a REPL interface, session logging, and a \doctor\ command to validate the local WaveTone installation. Real backend smoke tests are gated behind an explicit opt-in environment variable to ensure CI stability without a Windows host.

wavetone · high confidence

Adds the \cli-anything-wiremock\ Python CLI tool, which wraps the WireMock Admin REST API to let users manage HTTP stub mappings, inspect served requests, control stateful scenarios, and record real backend traffic directly from the terminal. The package includes core modules for stubs, requests, scenarios, recordings, and settings, along with a unified REPL skin for consistent terminal branding and output formatting. Comprehensive unit and end-to-end tests are provided to verify CLI argument parsing, JSON output modes, and live server interactions.

wiremock · high confidence

Added the core Python modules for the Zoom CLI harness, providing the foundational capabilities to interact with the Zoom API. The new \auth.py\ module implements an OAuth2 flow (including a local callback server and manual code entry) to handle user login and token management. \meetings.py\ exposes CRUD operations for scheduling, updating, and deleting meetings, along with retrieving join URLs. \participants.py\ manages meeting registrants (adding, listing, removing) and lists past meeting attendees. Finally, \recordings.py\ enables listing, downloading, and deleting cloud recordings, with the download function specifically handling direct URL retrieval with bearer token authentication.

_zoom/agent-harness/cli\anything/zoom/core · high confidence

Added a new Python package for the Zoom agent harness, providing a CLI tool to manage Zoom meetings via the Zoom REST API using OAuth2. The package, named cli-anything-zoom (version 1.0.1), includes console scripts for command-line interaction and depends on click, requests, and prompt-toolkit. It is distributed as a namespace package within cli\_anything and includes skill documentation files.

zoom/agent-harness · high confidence

Adds a new Zoom integration harness within the cli-anything framework, providing a unified terminal interface (ReplSkin) for consistent branding and user experience across CLI tools. The backend securely manages Zoom OAuth2 authentication by storing tokens and configuration files with owner-only permissions (0o600/0o700) on both Unix and Windows systems, and handles automatic token refresh for API requests.

_zoom/agent-harness/cli\anything/zoom/utils · high confidence

Added a Python-based CLI harness (cli-anything-quietshrink) that wraps the standalone quietshrink bash tool, providing structured JSON output for AI agents. The harness exposes commands to compress screen recordings with quality presets (tiny, balanced, transparent, pristine), probe video metadata, list presets, and verify the environment (ffmpeg, hevc\_videotoolbox, bash CLI). It includes 15 unit tests covering command wiring, JSON schema validation, and error handling, along with documentation and an agent skill definition.

quietshrink · high confidence

This change adds the core browser automation module, exposing filesystem-like navigation (ls, cd, cat, grep) and page operations (open, reload, back, forward) via the @apireno/domshell backend. It enforces security by validating URLs before navigation and raising a ValueError for invalid or dangerous schemes, and ensures session state (working directory and URL) is updated only when backend commands succeed.

_browser/agent-harness/cli\anything/browser/core · high confidence

Adds a new browser automation CLI harness (\browser/agent-harness\) that maps Chrome's Accessibility Tree to a virtual filesystem, allowing agents to use familiar shell commands like \ls\, \cd\, \cat\, and \grep\ for browser interaction. The harness integrates with the \@apireno/domshell\ 2.0.0+ MCP server, supporting both standard command-per-process execution and a persistent daemon mode for lower latency. It includes a Python package setup (\setup.py\) defining the \cli-anything-browser\ entry point and dependencies, along with comprehensive documentation (\HARNESS.md\) detailing the architecture, state management, and path syntax.

browser/agent-harness · high confidence

Added new utility modules in the browser harness: a DOMShell MCP client wrapper for browser automation, a unified REPL skin for consistent CLI branding, and security utilities that validate URLs (blocking dangerous schemes like file:// and javascript://, and optionally private networks) and sanitize DOM text to mitigate prompt injection.

_browser/agent-harness/cli\anything/browser/utils · high confidence

Adds a new command-line interface for AdGuardHome, allowing users to manage filtering rules, DNS rewrites, client devices, DHCP settings, query logs, and server status via the AdGuardHome REST API. The tool supports interactive REPL mode, structured JSON output for automation, and persists connection configuration (host, port, credentials, HTTPS flag) to a local JSON file with environment variable overrides. It includes a comprehensive test suite with unit tests mocking HTTP calls and E2E tests against a Dockerized AdGuardHome instance.

adguardhome · high confidence

Adds a new command-line interface for browser automation that maps Chrome's Accessibility Tree to a virtual filesystem via the DOMShell MCP server. Users can now navigate web pages and interact with elements using familiar shell-like commands (ls, cd, cat, grep, click, type) in both one-shot and interactive REPL modes. The tool supports JSON output for machine-readable results, daemon mode for persistent connections to improve performance, and includes safeguards such as skipping dependency checks for help/version flags and redacting sensitive text from action outputs.

_browser/agent-harness/cli\anything/browser · high confidence

Adds a new CLI harness for the Calibre e-book manager, providing a structured interface for library operations, metadata editing, and format conversion. The tool wraps the real Calibre backend tools (\calibredb\, \ebook-convert\, \ebook-meta\) and exposes them via a Click-based command-line interface and an interactive REPL. Key capabilities include connecting to a library, listing and searching books, editing metadata, managing custom columns, converting formats, and exporting books. The implementation includes core modules for session management, library operations, metadata handling, format conversion, and export pipelines, along with comprehensive unit and smoke tests.

_calibre/agent-harness/cli\anything · high confidence

A new Python package, cli-anything-qgis, is added to the QGIS agent harness to provide a stateful command-line interface for interacting with the real QGIS runtime. The harness splits responsibilities between PyQGIS for stateful project, layer, feature, and layout authoring, and the \qgis\_process --json\ CLI for processing algorithms and layout exports. It supports creating and managing \.qgs\/\.qgz\ projects, creating GeoPackage-backed vector layers, adding features via WKT, authoring print layouts, and exporting layouts to PDF or image. The tool exposes a stable JSON output format for all commands and includes a stateful REPL mode for interactive sessions, along with a session model for tracking project state and command history.

QGIS · high confidence

Adds a new CLI harness that provides programmatic access to Pioneer Rekordbox 6/7 by combining direct SQLCipher writes to the encrypted master.db library (via pyrekordbox) with live-deck control via virtual MIDI (using mido and the bundled Bunker.midi.csv mapping). Users can now inspect the library, create and modify playlists with automatic database backups, and control deck transport, EQ, and crossfading through a command-line interface designed for agent consumption.

rekordbox · high confidence

The eth2-quickstart area now includes a new \cli-anything-eth2-quickstart\ package that wraps the upstream \eth2-quickstart\ shell scripts with a structured Python CLI. This harness provides commands to orchestrate node setup (phase 1, phase 2, or auto), install execution/consensus/MEV clients, configure validator metadata (without touching secrets), expose RPC via Nginx or Caddy, and check health/status. It supports machine-readable JSON output, interactive REPL mode, and requires explicit \--confirm\ flags for mutating operations. The package includes unit and E2E tests, a SKILL.md definition for agent discovery, and a unified REPL skin for consistent terminal branding.

eth2-quickstart · high confidence

Adds a new \cli-anything-notebooklm\ package that wraps the community \notebooklm-py\ CLI to provide a structured interface for AI agents. This harness exposes command groups for authentication (\auth\), notebook management (\notebook\), source ingestion (\source\), chat (\chat\), artifact generation (\artifact\), downloads (\download\), and sharing (\share\). It includes a local session persistence layer to track the active notebook, enforces JSON output only for upstream commands that support it, and sanitizes sensitive authentication paths in error messages. The package is marked as experimental and unofficial, requiring a local Google-authenticated session and Python 3.10+.

notebooklm · high confidence

Adds a new \cli-anything-iterm2\ command-line tool that wraps the iTerm2 Python API, enabling AI agents and users to programmatically manage terminal windows, tabs, and sessions. This harness provides commands to send text, read screen contents, split panes, manage broadcast domains, invoke menu items, and handle tmux integration, all structured with JSON output for machine readability.

iterm2 · high confidence

Adds a new command-line interface for managing n8n workflow automation, enabling users to list, create, update, and delete workflows, executions, credentials, variables, and tags directly from the terminal or an interactive REPL. The tool includes built-in workflow scaffolds for common patterns (webhooks, API integrations, scheduled tasks), local version tracking with SQLite for rollback capabilities, and offline expression validation. It is built on the n8n Public API v1.1.1 and supports JSON output for machine-readable integration.

n8n · high confidence

Added a new sketch harness in sketch/agent-harness that converts JSON design specifications into .sketch files. The CLI (cli.js) provides build and list-styles commands, while the core builder (builder.js) handles token resolution, layout computation (layout.js), and Sketch layer generation (primitives.js). The implementation includes robust path validation with realpath canonicalization and symlink protection to prevent arbitrary file access when resolving token files, and uses a custom UTF-8 serialization method to avoid CJK character corruption in the output ZIP.

sketch/agent-harness · high confidence

The Live2D CLI harness has expanded from a read-only inspector to a full management tool, introducing 42 commands across editing, generation, analysis, and workflow categories. Users can now modify motion fade times, add or remove motions and expressions, replace texture paths, and edit model references, with all changes protected by automatic backups and an undo system. New capabilities include linting for best practices, detailed diffing between models, HTML snapshot generation for previews, batch editing of multiple models, and utilities for orphan file detection, texture atlas analysis, and format migration.

live2d · high confidence

A new command-line interface for FreeCAD has been added, enabling headless 3D modeling, assembly, CAM, and FEM workflows via Python macros executed through \freecadcmd\. The harness targets FreeCAD 1.1, introducing support for the new \LocalCoordinateSystem\ datum system, Whitworth threading, and inline assembly constraints. Users can now create, modify, and export CAD documents (STEP, STL, etc.) using a JSON-based state model and a suite of CLI commands covering PartDesign, Draft, TechDraw, and simulation.

freecad · high confidence

Added a Python-based CLI harness (\cli-anything-intelwatch\) that wraps the Node.js \intelwatch\ tool via \npx\, allowing users to run competitive intelligence and OSINT commands (such as generating company profiles) directly from the terminal. The package includes a unified REPL skin for consistent terminal branding and output formatting, along with unit and end-to-end tests verifying argument forwarding and error handling.

intelwatch · high confidence

A new \cli-anything-nslogger\ command-line tool is introduced to parse, filter, export, and monitor NSLogger log data. Users can now read and filter \.rawnsloggerdata\ and \.nsloggerdata\ files by level, tag, thread, time range, and text content, and export results to text, JSON, or CSV. The tool also supports live monitoring of iOS/macOS applications via Bonjour or direct TCP/TLS connections, with output mirroring to local files or JSON Lines streams. This harness is designed to integrate NSLogger diagnostics into agent workflows and CLI pipelines.

nslogger · high confidence

A new command-line interface (cli-anything-eez-studio) has been added to automate EEZ Studio workflows. It allows users to create, inspect, and modify native .eez-project JSON files directly, including adding LVGL screens and widgets, and managing SCPI instrument command metadata. For build and export operations, the harness can invoke the real EEZ Studio Node.js backend (via the EEZ\_STUDIO\_SOURCE environment variable) to inspect project metadata and build LVGL simulator artifacts, while providing structured error messages when the backend is unavailable.

eez-studio · high confidence

This release introduces a suite of new CLI harnesses that allow AI agents to interact with a variety of desktop and web applications. The 3MF harness adds mesh geometry editing capabilities, including cylindrical hole detection and resizing. New harnesses are available for QGIS (geospatial analysis), Audacity (audio evaluation and regression testing), AdGuardHome (DNS filtering), AnyGen (code generation), Browser automation (DOMShell integration), iTerm2 (terminal session management), Joplin (note-taking with atomic file locking and undo/redo), JumpServer (IT asset management), and WireMock (HTTP mocking). Additionally, a skill generator plugin has been added to automatically produce SKILL.md documentation for these harnesses by extracting metadata from their CLI structures.

python · high confidence

The \cli-hub-matrix\ area now includes structured skill definitions for five major creative and productivity domains: 3D & CAD, game development, image & graphic design, knowledge & research, and video creation. Each domain is defined by a \SKILL.md\ file that maps specific user capabilities (such as mesh modeling, agent-driven playtesting, AI image generation, document authoring, and video assembly) to a matrix of available providers, including CLI harnesses, Python libraries, native binaries, and cloud APIs. These files provide installation instructions, preflight checks to verify local tool availability, and provider selection constraints to help agents choose the best tool for a given task based on quality, cost, and offline requirements.

cli-hub-matrix · high confidence

Introduces the \cli-anything-sbox\ Python package (v1.0.0) in the \sbox/agent-harness\ directory, providing a command-line interface to manage s&box game engine projects without the GUI. This harness allows users to create and lint projects, manipulate scene and prefab JSON files (add/remove objects, components, and properties), generate C\# component and editor menu code, and manage configuration files for inputs, collisions, materials, sounds, and localization. It also includes utilities for asset listing, reference tracking, and dedicated server management.

sbox · high confidence

The Shotcut CLI harness now supports live preview generation via a new \preview.py\ module that renders low-resolution clips and sampled frames using a configurable recipe system. It introduces audio automation features, including \set\_volume\_envelope\ and \duck\_volume\ in \filters.py\, allowing precise control over clip volume over time. The XML parsing backend has been hardened by replacing \lxml\ with the standard library \xml.etree.ElementTree\ and routing untrusted XML through \defusedxml\ to prevent security vulnerabilities. Additionally, session state is now saved with atomic file locking to prevent race conditions, and the transition model has been updated to use Shotcut's sub-tractor format for better timeline compatibility.

_shotcut/agent-harness/cli\anything/shotcut/core · high confidence

The cli-hub package has been updated to version 0.4.1, introducing a new CLI-Matrix system that allows users to discover, install, and manage curated workflow matrices alongside standard CLIs. This update also migrates analytics tracking to PostHog, enabling the collection of usage data for these new matrix features, and includes a fix to ensure UTF-8 output compatibility on Windows systems with non-UTF-8 console encodings.

_cli-hub/cli\hub · high confidence

The Shotcut agent harness now uses defusedxml for untrusted XML parsing to address security vulnerabilities, and includes the necessary defusedxml dependency in setup.py. Additionally, the package now includes SKILL.md files via package\_data for pip install support, and the repository URL has been corrected to the official HKUDS/CLI-Anything location.

shotcut/agent-harness · high confidence

This update addresses a security vulnerability by adding the defusedxml library to handle untrusted XML parsing safely. It also corrects the project URL to point to the official HKUDS/CLI-Anything repository and bumps the version to 1.0.1. Additionally, the package configuration now includes Markdown skill files in the distribution to support pip install usage.

inkscape/agent-harness · high confidence

The Audacity CLI now automatically persists project changes after one-shot commands (such as adding tracks, clips, or effects) unless the session is in REPL mode or the new \--dry-run\ flag is used. This ensures that modifications are saved to disk immediately after execution, while \--dry-run\ allows users to preview command effects without altering the project file. Additionally, an evaluation harness has been added to the README, providing commands to run lightweight evaluation suites and generate reports for comparing results against baselines.

_audacity/agent-harness/cli\anything/audacity · high confidence

The Blender harness now supports object parenting in generated scripts, ensuring hierarchical relationships are preserved during scene creation. It also corrects the EEVEE render engine mapping to use the standard 'BLENDER\_EEVEE' enum instead of the deprecated 'BLENDER\_EEVEE\_NEXT', and fixes Windows path handling in render output scripts by removing raw-string prefixes that caused escape issues. Additionally, a new preview bundle system has been introduced to manage cached preview manifests and trajectories, while the REPL skin now displays skill installation commands and paths in the startup banner to aid agent discovery.

_blender/agent-harness/cli\anything/blender/utils · high confidence

The Kdenlive harness now enforces a strict allowlist of video and audio codecs (e.g., libx264, aac) and blocks extra arguments that could bypass these restrictions, preventing potential injection issues. Additionally, the MLT XML generation has been rewritten to produce Kdenlive Gen 5 (doc version 1.1) compatible output, ensuring native compatibility with modern Kdenlive versions. The REPL banner has also been updated to display skill installation commands and global skill paths for AI agents.

_kdenlive/agent-harness/cli\anything/kdenlive/utils · high confidence

This update introduces several improvements to the Shotcut harness utilities. Security is strengthened by validating video and audio codecs against a strict allowlist in the melt backend and blocking extra arguments that could bypass these restrictions, while MLT XML parsing now uses defusedxml to prevent unsafe external entity processing. A new preview bundle system is added, providing a caching mechanism for preview assets based on content fingerprints and protocol versions. The CLI interface is enhanced with a more detailed startup banner that displays skill installation commands and paths, and the REPL now auto-detects skill files from the repository root or package directory.

_shotcut/agent-harness/cli\anything/shotcut/utils · high confidence

The GIMP CLI harness now supports a --dry-run flag to execute commands without persisting changes to disk, and automatically saves the project state after one-shot commands if modifications were made. Text drawing on layers has been refactored to record operations in a persistent 'draw\_ops' history rather than overwriting layer properties, and the REPL now uses shlex for robust parsing of quoted strings with spaces. Additionally, the layer set command now accepts extra arguments to handle multi-word values.

_gimp/agent-harness/cli\anything/gimp · high confidence

The GIMP agent harness now lists Pillow as an optional extra rather than a hard requirement, allowing users to install the package without the heavy image-processing library if they do not need it. The project URL has been corrected to the official HKUDS/CLI-Anything repository, and the description now recommends GIMP instead of requiring it. Additionally, Markdown skill files are now included in the package distribution via package\_data, ensuring they are available when installed via pip.

gimp/agent-harness · high confidence

The GIMP image processing backend now uses GIMP-native Script-Fu commands (such as \gimp-file-load\ and \gimp-file-save\) instead of relying on external tools like Pillow, enabling direct rendering within the GIMP environment. To prevent command injection via file paths, all input and output paths passed to Script-Fu are now escaped. Additionally, the REPL skin has been updated to display skill installation commands and global skill paths in the startup banner, and the backend now exposes a map of CLI filter names to their corresponding GIMP Script-Fu equivalents.

_gimp/agent-harness/cli\anything/gimp/utils · high confidence

The setup script now detects Windows Bash environments (Cygwin, MinGW, MSYS) and verifies the presence of the 'cygpath' utility. If 'cygpath' is missing, the script terminates with a clear error message instructing users to install Git for Windows or use WSL, preventing cryptic path-resolution errors later in the process.

cli-anything-plugin/scripts · high confidence

The audio utilities now correctly encode 8-bit WAV files as unsigned PCM (0–255 range) instead of signed, ensuring accurate audio playback for this bit depth. The REPL startup banner has been expanded to display metadata including the command to install the relevant AI skill and the path to the global skill file, with paths displayed relative to the home directory when possible. Additionally, a new utility module provides a safe JSON writer with file locking support for Unix systems.

_audacity/agent-harness/cli\anything/audacity/utils · high confidence

The Kdenlive and OBS Studio command-line interfaces now support a --dry-run flag to execute commands without persisting changes to disk, and automatically save the project state after one-shot commands if modifications were made. Additionally, input parsing in the interactive REPL mode has been updated to use shlex.split for more robust handling of quoted arguments, falling back to simple splitting only when necessary.

_kdenlive/agent-harness/cli\_anything/kdenlive, obs-studio/agent-harness/cli\_anything/obs\studio · high confidence

The CLI-Anything Blender harness package has been refactored to support pip installation more robustly by moving skill documentation (SKILL.md) into the Python package and including it via package\_data. The setup script now uses pathlib for safer file handling, updates the project URL to the official HKUDS/CLI-Anything repository, and bumps dependency versions for click (\>=8.1) and prompt-toolkit (\>=3.0). These changes improve installation clarity and ensure skill files are correctly bundled with the package.

blender/agent-harness · high confidence

The session management now uses atomic file locking when saving project data, preventing potential race conditions and data corruption during concurrent writes. Additionally, the filter registry has been updated to include specific Kdenlive names for fade and speed effects (such as 'fade\_from\_black', 'fade\_to\_black', 'fadein', 'fadeout', and 'speed'), ensuring these transitions and effects are correctly identified and compatible with native Kdenlive projects.

_kdenlive/agent-harness/cli\anything/kdenlive/core · high confidence

The session save mechanism now uses atomic file locking to prevent race conditions when multiple processes access project files, and JSON keys are sorted to ensure deterministic output for agent diffing. Additionally, source transforms (position, size, crop, and rotation) are now validated using dedicated functions that reject non-finite numbers and enforce positive dimensions, preventing invalid geometry from being applied to sources.

_obs-studio/agent-harness/cli\_anything/obs\studio/core · high confidence

The LibreOffice CLI harness now supports importing existing Office and ODF files (DOCX, XLSX, PPTX, ODT, etc.) into editable project JSON via new \document import\ and \document import-formats\ commands, as well as enhanced \document open\ behavior for non-JSON inputs. A \--dry-run\ flag allows running commands without saving changes to disk, and auto-save is triggered after one-shot commands unless dry-run is active. Additionally, the REPL input parsing now uses \shlex.split\ for more robust argument handling, falling back to simple splitting on error.

_libreoffice/agent-harness/cli\anything/libreoffice · high confidence

The OBS Studio input validation utilities now explicitly reject non-finite numbers (NaN, Infinity) and boolean values in range checks, position coordinates, size dimensions, and crop parameters, raising clear errors instead of allowing invalid geometry. Additionally, the REPL skin has been updated to display a startup banner that includes AI skill installation instructions and the path to the global skill file, with improved path display logic that shows home-relative paths when possible.

_obs-studio/agent-harness/cli\_anything/obs\studio/utils · high confidence

SVG parsing functions in svg\_utils now use defusedxml to prevent XML-based security vulnerabilities when processing untrusted input. The REPL skin in repl\_skin.py has been updated to display a wider banner (72 characters) that includes metadata lines for the skill installation command and the global skill path, with paths displayed relative to the home directory when possible. The ReplSkin class now auto-detects SKILL.md files from either the repository root or the package directory, supporting agent discovery.

_inkscape/agent-harness/cli\anything/inkscape/utils · high confidence

The Shotcut CLI now supports a two-step media workflow where users must explicitly import files to obtain a clip\_id before adding them to the timeline, replacing the previous direct file-path approach. A new --dry-run flag allows testing commands without persisting changes, while an auto-save callback automatically writes the project to disk after each command when not in dry-run mode. The tool also introduces live preview capabilities, including background pollers and integration with cli-hub for real-time monitoring, alongside new filter commands for volume envelopes and ducking.

_shotcut/agent-harness/cli\anything/shotcut · high confidence

The Audacity harness now correctly renders tracks that contain only generator effects (such as silence or tone generation) even when no audio data is initially present, ensuring these effects are included in the final mix. Additionally, session files are now saved atomically using exclusive file locking, which prevents data corruption or race conditions when multiple processes access the project file simultaneously.

_audacity/agent-harness/cli\anything/audacity/core · high confidence

The CLI harnesses for Audacity, Kdenlive, and OBS Studio now point to the official HKUDS/CLI-Anything GitHub repository instead of placeholder URLs. Additionally, the skill documentation files (SKILL.md) are now included in the installed packages via package\_data, ensuring they are available when the harness is installed via pip.

audacity/agent-harness, kdenlive/agent-harness, obs-studio/agent-harness · high confidence

The cli-hub PyPI package (cli-anything-hub) has been updated to version 0.4.1, introducing a preview viewer for inspecting and rendering CLI harness sessions via commands like \cli-hub previews inspect\ and \html\. The package now includes vendored matrix skill content for offline access and supports Python 3.10 through 3.13. Analytics tracking has been standardized to use PostHog by default, with a legacy Umami option available via environment variable, and users can opt out using \CLI\_HUB\_NO\_ANALYTICS=1\.

cli-hub · high confidence

The Inkscape harness now correctly renders multi-line text with proper wrapping, line-height, and horizontal anchoring (start/middle/end) in both SVG export and Pillow-based image export, replacing the previous single-line text handling. Star shapes are no longer incorrectly marked with a non-standard 'sodipodi:type' attribute, ensuring they export as standard SVG paths. Additionally, project session files are now saved atomically using file locking to prevent data corruption from race conditions, and JSON keys are sorted for deterministic output.

_inkscape/agent-harness/cli\anything/inkscape/core · high confidence

The LibreOffice backend now uses a comprehensive set of headless flags to prevent interactive UI interference and includes a macOS-specific fallback via LaunchServices to resolve SIGABRT crashes during conversion. Additionally, XML parsing in ODF validation has been switched to defusedxml to prevent XXE attacks, and the REPL skin now auto-detects and displays skill paths in the startup banner.

_libreoffice/agent-harness/cli\anything/libreoffice/utils · high confidence

The GIMP harness now prioritizes the native GIMP Script-Fu backend for rendering projects, falling back to Pillow only when GIMP is unavailable. To eliminate the hard dependency on Pillow, the system now includes pure-Python header parsing for reading image dimensions (PNG, JPEG, GIF, BMP, WebP, TIFF) and basic metadata probing, ensuring the harness functions even without external image libraries. Additionally, session saves now use atomic file locking to prevent race conditions, and JSON output keys are sorted for deterministic diffing.

_gimp/agent-harness/cli\anything/gimp/core · high confidence

Added comprehensive test coverage for CLI Hub core modules; Added comprehensive test suite for Draw.io CLI harness; Added comprehensive test suite for Shotcut core and CLI workflows; Added test coverage for document import functionality; Added tests for Audacity evaluation harness and core functionality; Added tests for Blender script generation, object parenting, and live preview sessions; Added tests for SKILL.md discovery and installed package structure; Added tests for atomic file locking and draw operations; Added tests for non-finite value rejection in OBS geometry and range validators; Added tests for skill generator metadata extraction and SKILL.md generation; Added tests for text wrapping, star rendering, and document autosaving; Added tests for the sketch agent-harness build pipeline and security validations; Added unit and end-to-end test suites for the Zoom CLI harness; Added unit and end-to-end tests for the ComfyUI CLI harness; Updated end-to-end tests for Kdenlive XML generation.

This release introduces new agent harnesses for several platforms, providing CLI tools to interact with them. The JumpServer harness (v0.1.0) offers stateful management for bastion hosts, while the n8n harness (v2.4.7) enables workflow automation via the n8n API. A new Sketch harness (v1.0.0) allows generating Sketch files from JSON design specs. Additionally, harnesses are added for Uni-Mol Tools (molecular property prediction), Zotero (desktop integration), and NSLogger (log access). These new packages define their own dependencies, such as \click\, \requests\, \sketch-constructor\, and \commander\, and are structured as standalone CLI applications.

(dependencies) · medium confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 35 → 48 (+13.5)
• Rubric changed (rubric-2026.08.15 → rubric-2026.09.15) — scores are not directly comparable.

• Code Health 29 → 35 (+5.5)
• Architecture 97 → 100 (+3.3)
• Maturity 75 → 81 (+6.1)
• Readiness 22 → 44 (+21.8)
• Security 44 → 65 (+20.8)

• Context/problem and consequences/trade-offs are absent in the visible text (unimol_tools/agent-harness/docs/guides/02-QUICK-START.md)
• Coverage not measured — test suite did not build
• Decision/consequences are absent (no trade-offs or alternative paths for each problem) (unimol_tools/agent-harness/docs/guides/05-TROUBLESHOOTING.md)
• Dimension evaluation failed
• Disclosure policy has no reporting contact
• Duplicated block (10 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Formatting.cs)
• Duplicated block (10 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Formatting.cs)
• Duplicated block (10 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Formatting.cs)
• Duplicated block (11 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Actions.cs)
• Duplicated block (11 lines × 4) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Formatting.cs)
• Duplicated block (12 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Actions.cs)
• Duplicated block (12 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.MultiplayerState.cs)
• Duplicated block (13 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Actions.cs)
• Duplicated block (14 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.MultiplayerState.cs)
• Duplicated block (14 lines × 3) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Actions.cs)
• Duplicated block (15 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.MultiplayerState.cs)
• Duplicated block (15 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.StateBuilder.cs)
• Duplicated block (16 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.cs)
• Duplicated block (18 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.cs)
• Duplicated block (19 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.MultiplayerState.cs)
• …and 76 more

• Coverage not measured — .NET and JavaScript/TypeScript suite
• Dependency hygiene PARTLY measured — Python dependencies read, no exact pin to grade for currency
• Duplicated block (10 lines × 2) (anygen/agent-harness/cli_anything/anygen/utils/anygen_backend.py)
• Duplicated block (10 lines × 2) (browser/agent-harness/cli_anything/browser/browser_cli.py)
• Duplicated block (10 lines × 2) (freecad/agent-harness/cli_anything/freecad/core/draft.py)
• Duplicated block (10 lines × 2) (kdenlive/agent-harness/cli_anything/kdenlive/core/filters.py)
• Duplicated block (10 lines × 2) (libreoffice/agent-harness/cli_anything/libreoffice/core/calc.py)
• Duplicated block (10 lines × 2) (libreoffice/agent-harness/cli_anything/libreoffice/core/impress.py)
• Duplicated block (10 lines × 2) (libreoffice/agent-harness/cli_anything/libreoffice/utils/odf_utils.py)
• Duplicated block (10 lines × 2) (macrocli/agent-harness/cli_anything/macrocli/backends/gui_macro.py)
• Duplicated block (10 lines × 2) (macrocli/agent-harness/cli_anything/macrocli/macrocli_cli.py)
• Duplicated block (10 lines × 2) (n8n/agent-harness/cli_anything/n8n/n8n_cli.py)
• Duplicated block (10 lines × 2) (nsight-graphics/agent-harness/cli_anything/nsight_graphics/utils/backend/commands.py)
• Duplicated block (10 lines × 2) (sbox/agent-harness/cli_anything/sbox/core/localization.py)
• Duplicated block (10 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Formatting.cs)
• Duplicated block (10 lines × 2) (slay_the_spire_ii/agent-harness/bridge/plugin/BridgeMod.Formatting.cs)
• Duplicated block (10 lines × 2) (unimol_tools/agent-harness/cli_anything/unimol_tools/unimol_tools_cli.py)
• Duplicated block (10 lines × 24) (3MF/agent-harness/cli_anything/threemf/threemf_cli.py)
• Duplicated block (10 lines × 3) (docs/scripts/blender_orbital_relay_drone_demo.py)
• Duplicated block (10 lines × 3) (libreoffice/agent-harness/cli_anything/libreoffice/utils/odf_utils.py)
• …and 1297 more

• 22 commits — 13 feature/other, 9 fixes

• (root) — 6 commits
• (repo) — 4 commits
• adguardhome/agent-harness — 2 commits
• docs/hub — 2 commits
• inkscape/agent-harness — 2 commits
• .github/scripts — 1 commit
• audacity/agent-harness — 1 commit
• cli-anything-plugin/skill_generator.py — 1 commit
• cli-hub/cli_hub — 1 commit
• cursor-plugin/scripts — 1 commit
• eth2-quickstart/agent-harness — 1 commit

• fix: fix(audacity): render generators on empty tracks (#421)
• fix: fix(cli-hub): reconfigure stdout/stderr to UTF-8 on GBK consoles (Chinese Windows crash) (#419)
• fix: fix(inkscape): autosave new documents to project path (#430)
• fix: fix(inkscape): render stars as standard SVG paths (#442)
• fix: fix(skill-generator): parse nested decorator arguments (#444)
• fix: fix: add main guard to adguardhome CLI module (#438)
• fix: fix: add main guards to eth2-quickstart, mubu, siyuan and zotero CLIs (#440)
• fix: fix: normalize PyPI requirements in registry date updater (#446)
• fix: fix: report backend errors without a traceback and keep --json parsable (#439)
• change: Add Cursor generator plugin with slash commands and installers. (#417)
• change: Add Vivideo CLI to public registry (#425)
• change: Merge pull request #478 from HKUDS/chaohuang-ai-patch-3
• change: Merge pull request #479 from HKUDS/chaohuang-ai-patch-4
• change: Merge pull request #480 from HKUDS/chaohuang-ai-patch-5
• change: Merge pull request #481 from HKUDS/chaohuang-ai-patch-6
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: Update README.md
• change: feat(registry): add X Twitter Scraper CLI (#445)
• …and 2 more

• Containers 0 added · 0 removed · contexts 1 added · 0 removed · edges 0 added · 0 removed

• python

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

HKUDS/CLI-Anything was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.