danielmiessler/Fabric
51.8
Adequate · 13 November 2026
26.4k
lines of production code
Go
with TypeScript, Python
8
measurements over time
What this system is
Features
Add Amazon Bedrock AI vendor support with flexible authentication
Users can now connect to Amazon Bedrock models via a new plugin. The setup process dynamically fetches available AWS regions and supports three authentication modes: simple Bearer token (ABSK) authentication, explicit AWS access/secret keys, or the standard AWS credential provider chain. The plugin implements the Vendor interface to integrate with the existing AI framework.
internal/plugins/ai/bedrock · high confidence
Add Claude Code CLI provider with subscription authentication
Users can now route AI requests through the local Claude Code CLI, leveraging their existing Claude subscription login instead of requiring an API key. This new provider executes the \claude\ binary in an isolated temporary directory to prevent access to the host working folder, strips \ANTHROPIC\\\ environment variables to ensure billing goes to the subscription, and supports streaming responses via JSON output. It also handles image attachments by decoding data URLs into temporary files and passing them via \--add-dir\, while correctly mapping system messages and multi-part content to the CLI's arguments.
internal/plugins/ai/claudecode · high confidence
Add DigitalOcean Gradient AI Agents vendor support
Users can now connect to DigitalOcean's Gradient AI platform as a new AI vendor. The implementation adds a client that supports model listing via the DigitalOcean control plane API (using a specific token) or fallback to standard inference endpoints, and integrates i18n for error messages.
internal/plugins/ai/digitalocean · high confidence
Add Exolab AI plugin with secure URL validation
Users can now configure the Exolab AI plugin, which integrates with the OpenAI-compatible API structure. The setup process prompts for an API key, a base URL (defaulting to localhost), and a list of deployed models. A key behavioral change is the enforcement of secure connections: the plugin validates the base URL and will refuse to connect to remote hosts over plaintext HTTP, ensuring that API calls to external endpoints use HTTPS while allowing local development over HTTP.
internal/plugins/ai/exolab · high confidence
Add Firecrawl web search tool
Users can now perform web searches via the Firecrawl API by using the --firecrawl\_search flag. This new tool queries the Firecrawl search endpoint, scrapes the top five web results into markdown format, and returns them as structured sections. An API key is required for authentication, and the client includes a 90-second timeout to accommodate Firecrawl's scraping process.
internal/tools/firecrawl, internal/tools/serply · high confidence
Add Microsoft 365 Copilot integration
Users can now connect to Microsoft 365 Copilot as a new AI vendor. This integration requires an Azure AD app registration with specific delegated permissions and supports configuration via tenant ID, client ID, and OAuth2 credentials. The plugin exposes a single unified model and handles conversation creation and message streaming through the Microsoft Graph API.
internal/plugins/ai/copilot · high confidence
Add Nix package definition for Fabric AI
This change introduces the Nix build infrastructure for the Fabric AI application, enabling users to install and run the tool via the Nix package manager. The package definition (default.nix) builds the Go-based application from source, pinning the version to 1.4.480 and configuring the build to prevent automatic Go toolchain downloads. It also ensures that shell completion scripts for Zsh, Bash, and Fish are installed alongside the binary.
nix/pkgs · high confidence
Add OpenAI Codex vendor with browser-based OAuth authentication
Users can now connect to the OpenAI Codex backend using a new vendor integration. This adds browser-based OAuth login (with PKCE) to authenticate and manage access tokens, automatically refreshing them and persisting them securely. The integration supports streaming responses, handles provider-specific error messages (such as usage limits and login failures) with localized user feedback, and enforces secure HTTPS connections for all API and authentication endpoints.
internal/plugins/ai/codex · high confidence
Add Spotify podcast metadata retrieval tool
Users can now retrieve metadata for Spotify shows and episodes by providing a Spotify URL. This new tool in internal/tools/spotify integrates with the Spotify Web API using OAuth2 Client Credentials flow, requiring a Client ID and Client Secret configured via environment variables. It supports extracting metadata from show and episode URLs, searching for shows, and listing episodes for a given show. The tool also includes comprehensive unit and integration tests to verify URL parsing, metadata retrieval, and formatting functionality.
internal/tools/spotify · high confidence
Add custom patterns plugin with localized setup
A new internal tool plugin for managing custom patterns has been added. It introduces a setup flow that allows users to specify a directory for their custom patterns, automatically expanding home-directory shortcuts (e.g., \~/) and creating the directory if it does not exist. All user-facing strings, including the setup description and configuration prompts, are now localized via the i18n system rather than using hardcoded English text.
_internal/tools/custom\patterns · high confidence
Add internationalization (i18n) support with automatic locale detection and fallback
The application now supports multiple languages, including English, Spanish, Arabic, Portuguese (with distinct Brazilian and European variants), German, French, Japanese, Chinese, Turkish, and Persian. The system automatically detects the user's preferred language from environment variables (LC\_ALL, LC\_MESSAGES, LANG) and falls back to English if the locale is unavailable. It also supports downloading missing translation files from GitHub and caching them locally, ensuring that new or regional locales are available even if not embedded in the binary.
internal/i18n · high confidence
Add shell completions for Zsh, Bash, and Fish
Shell completions are now available for Zsh, Bash, and Fish, providing dynamic suggestions for patterns, models, contexts, sessions, vendors, strategies, extensions, Gemini TTS voices, and transcription models. The completions also support fixed-value flags (such as thinking levels, image dimensions, and debug levels) and file path completion for attachments, outputs, and configuration files. A setup script is included to automatically install these completions for the current shell.
completions · high confidence
Added example templates and scripts for the extension system
The Examples directory now includes a comprehensive guide and a set of ready-to-use templates demonstrating how to configure and use the extension system. These examples cover various scenarios, including a Python-based word generator, direct executable wrappers for SQLite3 and OpenAI, and shell scripts for generating local and remote security reports. The included YAML configurations and shell scripts illustrate how to define extension operations, handle input parameters, and manage output methods (stdout vs. file), providing users with concrete reference implementations for integrating external tools.
internal/plugins/template/Examples · high confidence
Anthropic client now supports Claude 5 models and structured thinking
The Anthropic plugin has been updated to support the latest Claude 5 family models (Fable, Sonnet, and Opus 5) and Opus 4.8, with automatic handling of their 1M context window requirements via beta headers. It also introduces support for structured thinking levels (Off, Low, Medium, High) and enforces parameter restrictions for newer models that reject non-default sampling settings. Additionally, the client now includes built-in web search tool support and comprehensive test coverage for model listing, parameter building, and search functionality.
internal/plugins/ai/anthropic · high confidence
Automated GitHub release description updates with fork support
The changelog tool now includes a new \--release\ flag that automatically updates the description of a GitHub release with the generated AI summary. This feature handles both standard repositories and forks; if the current repository is a fork, the tool attempts to update the release description in the upstream parent repository first, falling back to the fork itself if necessary.
_cmd/generate\changelog/internal · high confidence
Automated README feature section generation
Added a new Python script and documentation in the \scripts/readme\_updates\ directory to automate the generation of the 'Recent Major Features' section for the main README. The script queries the changelog SQLite database, applies keyword-based heuristics to filter for feature and improvement releases while excluding bug fixes and chores, and outputs formatted Markdown links. This allows maintainers to easily update the README with the latest significant product capabilities.
_scripts/readme\updates · high confidence
Cross-platform desktop notifications with secure command execution
Users can now receive desktop notifications on macOS, Linux, and Windows. The system automatically selects the best available backend (terminal-notifier, osascript, notify-send, or PowerShell) and uses environment variables to safely pass titles and messages, preventing shell injection attacks. On unsupported platforms, notifications are silently ignored.
internal/tools/notifications · high confidence
Expand OpenAI-compatible provider support with new vendors and configuration options
The OpenAI-compatible plugin now supports a significantly larger set of third-party AI providers, including new additions such as DemonRoute, TrustedRouter, Requesty, Infermatic, Eden AI, Z AI, OrcaRouter, Synthorai, Abacus, API Route, Y-API, Pzero, Apple Foundation Models, llmman, Cheaper Inference, Opper, FuturMix, and Novita AI. This update introduces flexible configuration capabilities, allowing providers to define custom model listing endpoints, support for the OpenAI Responses API (with specific tool name overrides for providers like xAI), and session-routing headers for OpenCode integration. It also adds support for providers that do not require an API key, such as local Apple Foundation Models, by validating the base URL instead. Additionally, MiniMax models now default to raw mode for correct message formatting, and the system includes improved error handling and direct model fetching for non-standard API responses.
_internal/plugins/ai/openai\compatible · high confidence
Initial Ollama plugin with streaming, multi-modal, and error-handling improvements
Users can now connect to local Ollama instances via a new plugin that supports streaming responses, multi-modal image inputs, and structured usage metadata. The plugin automatically converts single system-role messages to user-role to prevent empty responses, validates data URLs for images, and ensures the stream channel is closed on errors to prevent resource leaks.
internal/plugins/ai/ollama · high confidence
Introduce AI-powered changelog generation with incremental caching
The changelog generator now supports automated, AI-enhanced release notes. It uses incremental caching to speed up generation by reusing previously processed git history and PR data, only walking new commits since the last tag. The system aggregates incoming PR files and direct commits, correctly associating PR numbers with version tags and detecting merge commits via parent count and message patterns. For each version, it can invoke an external AI summarizer (defaulting to claude-opus-5 via the fabric CLI) to condense raw commit logs into concise, human-readable bullet points, while preserving PR headers and links. The generator also handles version date calculation based on the most recent commit in the PRs, ensures proper newline handling in the output file, and integrates with the incoming directory workflow for staged releases.
_cmd/generate\changelog/internal/changelog · high confidence
Introduce DryRun plugin for request simulation
A new DryRun plugin has been added to the AI plugin system, allowing users to simulate AI requests without sending them to a real provider. This plugin intercepts chat messages and options, formats them into a readable text representation (including model settings, temperature, and multi-content details), and returns a fake response with simulated token usage. It supports both streaming and non-streaming modes, enabling users to verify how their prompts and configurations would be structured before actual execution.
internal/plugins/ai/dryrun · high confidence
Introduce Fabric Pattern Studio UI with loopback-only binding
A new Streamlit-based user interface for the Fabric Pattern Studio is added, providing features such as model selection, pattern execution, and output management. To enhance security, the UI is configured to bind exclusively to the loopback address (127.0.0.1), ensuring it is not accessible from the network, and includes robust logging to a local file.
_scripts/python\ui · high confidence
Introduce GitHub changelog client with PR and commit data models
This change adds the internal GitHub client and data types used to fetch Pull Request and commit information for changelog generation. The new \client.go\ implements concurrent fetching of PRs with a semaphore-based limiter, distinguishes between lightweight validation queries and full commit retrieval, and handles GitHub authentication via OAuth2 tokens. The \types.go\ file defines the \PR\, \PRDetails\, and \PRCommit\ structs, including support for author type classification (user, organization, bot) and email fields. Tests in \email\_test.go\ verify correct handling of commit email data.
_cmd/generate\changelog/internal/github · high confidence
Introduce automated pattern metadata extraction and management workflow
This change introduces a new system for managing pattern descriptions and tags, centered around the \extract\_patterns.py\ script. The script automatically scans the local patterns directory (e.g., \\~/.config/fabric/patterns\ or \data/patterns\), extracts the first 500 words of each pattern's \system.md\ file into \pattern\_extracts.json\, and creates initial placeholder entries in \pattern\_descriptions.json\. This workflow simplifies the process of adding new patterns by automating the creation of reference extracts and metadata structures, ensuring that new patterns are immediately recognized and ready for description and tagging updates. The system also includes documentation (\README\_Pattern\_Descriptions\_and\_Tags\_MGT.md\) explaining how to complete these metadata entries and maintain synchronization with the web interface.
_scripts/pattern\descriptions · high confidence
Introduce code2context CLI for scanning code projects into structured JSON
A new \code2context\ command-line tool is available to scan code directories or specific files and output a structured JSON representation of the project structure, file contents, and user-provided instructions. Users can run it in directory mode (scanning a folder up to a specified depth while ignoring patterns like \.git\ and \node\_modules\) or pipe a list of files via stdin. The output includes a hierarchical file tree, a report with file/directory counts, and the provided instructions, making it suitable for feeding context to AI agents. Symlinks pointing outside the project are safely skipped to prevent information leakage.
cmd/code2context · high confidence
Introduce git repository walker for changelog generation
Added new internal packages (types.go, walker.go) to handle git repository operations for changelog generation. The walker provides functionality to retrieve the latest tag, walk commits since a specific tag, and parse commit messages to extract version numbers and pull request numbers. This enables the changelog generator to accurately associate PRs with version tags and identify unreleased changes.
_cmd/generate\changelog/internal/git · high confidence
Introduce prompt strategy management with secure loading
The application now supports a new 'Prompt Strategies' plugin that allows users to manage and load custom prompt strategies from JSON files. This feature includes an automated setup process that can download a default set of strategies from a Git repository, configurable via environment variables for the repository URL and folder path. The implementation ensures security by validating file paths to prevent directory traversal attacks when loading strategy files, and it handles configuration validation to ensure strategies are properly loaded and available for use.
internal/plugins/strategy · high confidence
Introduce standalone changelog generator CLI with AI summaries and GitHub release updates
A new \generate\_changelog\ command is added to the CLI, providing a high-performance tool to generate comprehensive changelogs from git history and GitHub pull requests. It features a one-pass git walking algorithm, SQLite-based caching for incremental updates, and concurrent GitHub API calls (optimized with GraphQL) to fetch PR details. Users can now generate changelogs for specific versions or the full history, enable AI-powered summaries via Fabric, and use the \--release\ flag to automatically update GitHub release descriptions with AI summaries. The tool also supports processing incoming PR files, synchronizing the local database with git history and GitHub, and allows processing closed PRs.
_cmd/generate\changelog · high confidence
Introduce structured chat message model and streaming chatter implementation
The internal chat and core packages now use a structured \ChatCompletionMessage\ type that supports multi-part content (text and image URLs), tool calls, and reasoning content, with custom JSON marshaling to handle both legacy single-content and new multi-content formats. The \Chatter\ component has been rewritten to support structured streaming, including a \SuppressThink\ option to filter out AI reasoning blocks from the output, case-insensitive model name matching, and improved error handling for stream updates. The \PluginRegistry\ now initializes a comprehensive set of AI vendors (including Codex, VertexAI, Azure Entra ID, and Microsoft 365 Copilot) with proper configuration and token persistence, and includes tests for vendor selection, model prefix parsing, and think-block suppression. Also in this area: Security and audit improvements for server responses and configuration; Fixes for plugin configuration validation and server authentication security; Hardened server security, terminal output sanitization, and template safety; Firecrawl web search integration and --print-prompt validation fixes; Configurable CORS support for REST and Ollama servers.
internal/core · high confidence
Introduces SQLite-based caching for changelog generation
The changelog generator now persists processed data to a local SQLite database, caching versions, commits, pull requests, and metadata. This allows the tool to track the last processed tag and avoid re-processing historical data, improving performance and reliability for subsequent runs.
_cmd/generate\changelog/internal/cache · high confidence
Introduces centralized debug logging with granular verbosity levels
The application now uses a centralized logging package in internal/log to manage debug output, replacing scattered stderr prints. Users can control verbosity via a new --debug flag (or SetLevel API) with five levels: Off, Basic, Detailed, Trace, and Wire. The Wire level specifically enables full logging of LLM request/response exchanges, aiding in debugging model interactions. All debug messages are prefixed with 'DEBUG:' and written to stderr by default, while critical messages use the unconditional Log function.
internal/log · high confidence
New Azure AI Gateway plugin supporting Bedrock, Azure OpenAI, and Vertex AI
Users can now connect to multiple AI backends through a single Azure API Management gateway. The new plugin supports AWS Bedrock (Anthropic models), Azure OpenAI (GPT and o-series models), and Google Vertex AI (Gemini models). Configuration requires a gateway URL, a subscription key, and a backend type selection; Azure OpenAI also supports an optional API version (defaulting to 2025-04-01-preview). The plugin handles authentication, request formatting, and response parsing for each backend, including specific handling for sampling parameters (temperature/top\_p) on models that reject them.
internal/plugins/ai/azureaigateway · high confidence
New Nix shell and treefmt configurations with Go toolchain isolation
This change introduces two new Nix files: \nix/shell.nix\ and \nix/treefmt.nix\. The shell configuration provides a development environment with Go tooling (gopls, gotools, goimports-reviser, gomod2nix) and a custom \update-mod\ helper script to manage Go modules and regenerate Nix packages. Crucially, it sets the \GOTOOLCHAIN=local\ environment variable to prevent automatic Go toolchain downloads. The treefmt configuration enables formatting and linting tools (deadnix, statix, nixfmt) and wraps Go-specific formatters (goimports, gofmt) to also enforce the \GOTOOLCHAIN=local\ setting, ensuring consistent behavior across development workflows.
nix · high confidence
New PDF conversion and HTML readability tools
Users can now convert LaTeX content to PDF via the new \to\_pdf\ command-line tool, which accepts input from stdin or files and invokes \pdflatex\ to generate the output. Additionally, the \internal/tools/converter\ package introduces an \HtmlReadability\ function that extracts clean text content from HTML strings using the \go-readability\ library, accompanied by unit tests to verify its behavior.
_cmd/to\pdf, internal/tools/converter · high confidence
New REST API server with authentication and security hardening
The internal server now exposes a new REST API (powered by Gin) that consolidates chat, pattern, context, session, and configuration management into a single service. Authentication is enforced via an X-API-Key header for all endpoints except Swagger documentation, and binding to non-loopback addresses requires a key to be set. Security is improved by masking API keys in configuration responses, rejecting path-traversal attempts in storage and pattern names, and hiding internal filesystem paths from error messages. The chat endpoint supports streaming responses, pattern variables, and multi-turn sessions, while the configuration endpoint allows updating provider keys and URLs.
internal/server · high confidence
New Template System with Secure Extension Execution and Internationalization
The internal template plugin now provides a comprehensive system for variable substitution and dynamic content generation using a double-brace syntax (e.g., {{name}}). This update introduces a plugin architecture with built-in capabilities for datetime operations (e.g., {{plugin:datetime:now}}), file system access (e.g., {{plugin:file:read}}), and HTTP fetching (e.g., {{plugin:fetch:get}}). A key behavioral change is the introduction of a secure extension executor that shell-escapes all user-controlled values to prevent command injection when executing external scripts. Additionally, all user-facing strings and error messages have been extracted for internationalization (i18n) support, and the system now includes a registry for managing and validating external extension configurations.
internal/plugins/template · high confidence
New Vertex AI provider supporting dynamic model listing and Claude models
Added a new Vertex AI plugin that enables access to Anthropic Claude models via Google Cloud Vertex AI, alongside Google's Gemini models. The provider dynamically lists third-party models (like Claude) from the Model Garden API and includes a curated list of Gemini models, presenting them to users in a prioritized order (Gemini first, then Claude). It supports both standard and streaming chat interactions, requiring configuration of a Google Cloud Project ID and Region. Also in this area: 3 commits (0 fixes) modifying internal/plugins/ai; AI providers now respect caller context cancellation; Fix Gemini safety filter errors and Claude sampling parameter handling.
internal/plugins/ai/vertexai · medium confidence · unverified
New domain types and security hardening for file operations and streaming output
This change introduces several new domain-level capabilities and security controls. It adds structured streaming support via \StreamUpdate\ and \UsageMetadata\ types, and a \ThinkingLevel\ enum with token budgets for reasoning models. A new \ExtractFencedCodeBlock\ function allows extracting specific Markdown code blocks from text. Security is significantly hardened in the file manager: \ParseFileChanges\ and \ApplyFileChanges\ now strictly validate paths using \filepath.IsLocal\ and \os.OpenRoot\ to prevent arbitrary file writes and directory traversal, rejecting suspicious paths and control characters. Additionally, \SanitizeTerminalOutput\ removes dangerous terminal escape sequences (like clipboard writes and cursor control) from displayed text while preserving color codes, and \StripThinkBlocks\ uses a cached regex to efficiently filter out AI reasoning blocks.
internal/domain · high confidence
New maintenance audit script and Windows setup utility
Added a new \scripts/audit-patterns.sh\ tool that scans the patterns directory for maintenance issues, including thin or bloated pattern files, stale vendor model references (e.g., GPT-4, ChatGPT), missing input markers, i18n locale key gaps, and shell completion mismatches. Also introduced \scripts/setup\_fabric.bat\ to automate the installation of Fabric on Windows, handling prerequisites like Go and Git via Chocolatey, and prompting for API keys.
scripts · high confidence
One-line installer scripts for Fabric
Users can now install Fabric using a single command on Unix/Linux/macOS (via bash) and Windows (via PowerShell). The new scripts in the installer directory automatically detect the operating system and architecture, download the latest release binary, extract only the necessary executable, and place it in a specified directory (defaulting to \~/.local/bin or %USERPROFILE%\\.local\\bin). The installation process includes verification of the binary and provides clear instructions for updating the PATH if the installation directory is not already included.
scripts/installer · high confidence
OpenAI plugin refactored with Responses API support, model caching, and audio/image capabilities
The OpenAI plugin has been restructured into multiple focused files, introducing support for the new OpenAI Responses API alongside the existing Chat Completions API. Model discovery is now cached locally for 24 hours to reduce rate-limit errors, with a fallback to direct API fetching for non-standard providers. The plugin now supports audio transcription with parallel chunk processing and image generation via the Responses API, including tool-based workflows for search and image output. Streaming responses are handled with structured updates, and error messages are localized via i18n. Tests cover caching behavior, model fetching, audio validation, and image generation tool configuration.
internal/plugins/ai/openai · high confidence
Project migrates to Go with Nix build support and comprehensive documentation
The project has been rewritten in Go, replacing the previous Python implementation, and introduces a Nix flake for reproducible development and packaging (including a \fabric-slim\ variant and a full package bundling \yt-dlp\). To support this transition, the repository now includes a \.goreleaser.yaml\ configuration for building binaries across Linux, macOS, and Windows, alongside a complete overhaul of the documentation: the main \README.md\ and a new \README.zh.md\ (Chinese) provide updated installation guides, usage instructions, and feature lists reflecting the new architecture and supported AI providers.
(repo-wide) · high confidence
YAML configuration support and secure desktop notifications
Users can now persist CLI settings by providing a YAML configuration file via the new --config flag, with CLI arguments taking precedence over YAML values and defaults. The CLI also introduces cross-platform desktop notifications upon command completion, which can be customized via a --notification-command flag; this feature uses secure positional argument passing to prevent shell injection when executing custom notification scripts.
internal/cli · high confidence
Security
Introduce filesystem-backed storage with path traversal protection
The internal plugin database now uses a new filesystem-based storage implementation (fsdb) for patterns, sessions, and contexts. This change introduces strict validation for storage names, rejecting path traversal attempts (e.g., \..\, \/etc/passwd\) and Windows-specific reserved names (e.g., \CON\, \NUL\) to prevent unauthorized file access. It also adds atomic writes with exclusive file locking for \.env\ configuration updates to ensure data integrity during concurrent access, and supports custom pattern directories via the \CUSTOM\_PATTERNS\_DIRECTORY\ environment variable.
internal/plugins/db · high confidence
Behavioural changes
Azure OpenAI plugin now uses the openai-go SDK with Entra ID authentication support
The Azure AI plugin has been rewritten to use the official \openai-go/azure\ SDK, replacing the previous implementation. This change introduces a new \AzureEntra\ client variant that authenticates using Azure Entra ID (via \azidentity\) instead of API keys, while the standard Azure client continues to support API key authentication. Both clients now default to the \2025-04-01-preview\ API version if not explicitly configured and use a shared middleware to correctly inject deployment names into API request paths (e.g., \/openai/deployments/{name}/chat/completions\). The plugin also adds support for newer models like GPT-5 and o-series reasoning models, which are automatically detected to enable raw mode when necessary.
internal/plugins/ai/azure · high confidence
CLI error output redirected to stderr
The Fabric CLI now writes errors and streaming output to standard error (stderr) instead of standard output. This ensures that error messages do not interfere with command output when the CLI is used in scripts or piped to other tools, improving reliability for automation workflows.
cmd/fabric · high confidence
Case-insensitive vendor and model name matching
The AI plugin now handles vendor and model names in a case-insensitive manner, allowing users to specify vendors (e.g., "openai", "OpenAI") and models (e.g., "gpt-4o", "GPT-4O") without worrying about exact capitalization. The \VendorsManager\ stores and looks up vendors using lowercase keys, and the \VendorsModels\ component provides utilities to filter vendors and find specific model names regardless of case. This ensures that user input is matched correctly against the available AI providers and their models.
internal/plugins/ai · high confidence
Centralized GitHub token retrieval utility
A new utility function, GetTokenFromEnv, has been added to the changelog generation tool to standardize how GitHub authentication credentials are obtained. This function implements a specific precedence order for token resolution: it first checks a provided token argument, then falls back to the GITHUB\_TOKEN environment variable, and finally to the GH\_TOKEN environment variable. This change ensures consistent token handling across the tool's operations.
_cmd/generate\changelog/util · high confidence
Gemini plugin upgraded to new genai SDK with TTS, search, and thinking support
The Gemini plugin has been migrated to the new \google.golang.org/genai\ SDK, enabling dynamic model listing and support for the \gemini-2.5-flash-preview-tts\ model for text-to-speech audio output. Users can now enable web search with location validation, configure Gemini's thinking capabilities (low, medium, high, or custom token budgets), and receive responses with formatted source citations. The update also includes a curated list of selectable TTS voices and fixes for message role mapping and streaming stability.
internal/plugins/ai/gemini · high confidence
Git file fetching now uses a CLI fallback and displays localized error messages
The \FetchFilesFromRepo\ function in the \githelper\ tool now attempts to clone repositories using the \go-git\ library first, and automatically falls back to the system \git\ CLI if that fails, improving compatibility in environments where the Go library may struggle (such as DNS resolution issues on Termux). Additionally, all error messages generated during this process have been replaced with internationalization (i18n) translation keys, ensuring users see properly localized feedback in their supported language.
internal/tools/githelper · high confidence
Improved safety filtering and token accounting for Gemini responses
The Gemini integration now correctly handles safety blocks and image-specific content filters (prohibited content, recitation) by returning explicit errors instead of silent empty responses. Additionally, thought summaries generated during reasoning are excluded from the final output text, while their associated tokens are included in the output token count for accurate billing and usage reporting.
internal/plugins/ai/geminicommon · high confidence
Introduction of i18n error handling in attachment domain logic
The new \internal/domain/attachment.go\ file implements core attachment handling (ID generation, type resolution, content retrieval) and replaces hardcoded error strings with internationalization (i18n) translation keys. This ensures that error messages related to missing attachment content, unavailable files, or undetermined MIME types are now localized rather than hardcoded.
github.com/danielmiessler/fabric · high confidence
LM Studio plugin adds optional API key authentication and structured streaming
The LM Studio AI plugin now supports optional API key authentication, automatically attaching a Bearer token to requests when a key is configured, and implements structured streaming that exposes token usage metadata (input, output, and total tokens) alongside content updates during chat completions.
internal/plugins/ai/lmstudio · high confidence
New utility functions for path resolution and grouped item selection
The internal/util package now includes new helper functions to improve configuration handling and user interface consistency. GetAbsolutePath resolves file paths by expanding tilde (\~) and symlinks, while GetDefaultConfigPath locates the default configuration file at \~/.config/fabric/config.yaml with improved error handling. Additionally, a new GroupsItemsSelector utility provides case-insensitive sorting and selection for grouped items, ensuring consistent ordering in user-facing lists. Also in this area: Fixes for plugin configuration validation and server authentication security; Hardened server security, terminal output sanitization, and template safety.
internal/util · high confidence
Refactored CLI tooling with modular defaults, language settings, and robust pattern loading
The internal tools package has been restructured to support modular command handling and improved configuration management. A new defaults plugin now allows users to select and verify AI models via an interactive setup process, while a dedicated language plugin enables setting the output language with automatic tag validation. The patterns loader has been significantly enhanced to prevent temporary directory leaks by creating temp folders only during population and ensuring cleanup even on failure, and it now preserves custom user patterns when updating from the remote repository. These changes are backed by new tests verifying the safe handling of temporary files and configuration states.
internal/tools · high confidence
YouTube tool now supports optional API keys and configurable visual extraction
The YouTube tool has been refactored to make the YouTube API key optional, allowing users to extract transcripts without authentication while still supporting comments and metadata when a key is provided. Additionally, the tool now supports visual extraction using FFmpeg and Tesseract, with configurable CLI flags for parameters like concurrency and timeouts. The implementation also introduces smart subtitle language fallbacks, improved timestamp parsing (including fractional seconds), and better error handling for VTT parsing and duplicate segments.
internal/tools/youtube · high confidence
Fixes
Secure plugin endpoint validation and sensitive data masking
The plugin configuration system now enforces secure connections by rejecting plaintext HTTP URLs for remote hosts, allowing HTTP only for loopback addresses (localhost, 127.0.0.1, ::1) to support local development. Additionally, sensitive settings (identified by names containing KEY, TOKEN, SECRET, or PASSWORD) are automatically masked in setup prompts and output to prevent credential leakage. The setup process also allows users to reset required fields without triggering validation errors, and accepts carriage returns in interactive prompts for better compatibility.
internal/plugins · high confidence
Dependencies
Initial dependency manifests for Go backend and SvelteKit web frontend
The project now includes its first dependency manifests, establishing the baseline for the Go backend and the SvelteKit web frontend. The Go module (go.mod) defines the backend stack, requiring Go 1.26.0 and pinning key libraries such as the Anthropic SDK (v1.67.0), OpenAI Go SDK (v1.12.0), AWS SDK for Bedrock, and Ollama. The web frontend (web/package.json) initializes the client-side stack with Svelte 5, Vite 8, Tailwind CSS, and Shiki, while also introducing security overrides for several npm packages (e.g., qs, form-data, esbuild) to mitigate known vulnerabilities.
(dependencies) · high confidence
Web UI scaffold and configuration overhaul
The web application has been restructured with a new SvelteKit-based scaffold, introducing a modernized development and runtime environment. This includes a new \.browserslistrc\ file to enforce support for recent browser versions (Chrome/Firefox/Edge ≥89, Safari ≥15), a \pnpm-workspace.yaml\ to manage package resolution and enforce security overrides for vulnerable transitive dependencies (such as \qs\, \cookie\, and \esbuild\), and a \vite.config.ts\ extension that increases the proxy timeout to 15 minutes. The project now utilizes Tailwind CSS 4 for styling, configured via \app.css\ and \tailwind.config.ts\, and integrates the Skeleton UI library with a new set of theme imports (including 'rocket' and 'nouveau'). Additionally, the repository now includes a \dependabot.yml\ for automated weekly dependency updates, a \prettierrc\ for Svelte-aware code formatting, and an \eslint.config.js\ to standardize linting rules for the new stack.
web · high confidence
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
How this codebase got here
Score
- CAI 51 → 52 (+1.0)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.3) — scores are not directly comparable.
Lenses
- Code Health 64 → 80 (+16.3)
- Architecture 77 → 77 (-0.0)
- Maturity 80 → 80 (+0.5)
- Readiness 45 → 45 (+0.3)
- Security 72 → 68 (-4.1)
- Domain Modelling 53 → 53 (+0.0)
- Accessibility 48 → 49 (+0.7)
- Performance 80 → 100 (+20.0)
Resolved (24)
- Duplicated block (5 lines × 2) (scripts/readme_updates/update_readme_features.py)
- Edited copy of a member (30 corresponding lines) (internal/plugins/ai/anthropic/anthropic.go)
- FileTooLong: python_ui/streamlit.py (scripts/python_ui/streamlit.py)
- Hotspot: cmd/code2context/code.go (cmd/code2context/code.go)
- Hotspot: internal/plugins/ai/codex/codex.go (internal/plugins/ai/codex/codex.go)
- Medium CVE: GO-2026-4599 (go.mod)
- No ADRs found
- Off-boarding risk: anonymized user #1
- Outdated: google.golang.org/genai
- Rotate the exposed credentials — git history can't be un-committed
- Secret: jwt (client/config_example/config.yaml)
- extract_patterns.extract_pattern_info (cognitive 19) (scripts/pattern_descriptions/extract_patterns.py)
- streamlit.bulk_edit_patterns (cognitive 27) (scripts/python_ui/streamlit.py)
- streamlit.create_pattern (cognitive 30) (scripts/python_ui/streamlit.py)
- streamlit.create_pattern (cyclomatic 18) (scripts/python_ui/streamlit.py)
- streamlit.execute_patterns (cognitive 31) (scripts/python_ui/streamlit.py)
- streamlit.execute_patterns (cyclomatic 18) (scripts/python_ui/streamlit.py)
- streamlit.main (cognitive 253) (scripts/python_ui/streamlit.py)
- streamlit.main (cyclomatic 66) (scripts/python_ui/streamlit.py)
- streamlit.pattern_creation_wizard (cognitive 35) (scripts/python_ui/streamlit.py)
- …and 4 more
New (46)
- Critical CVE: [GHSA redacted] (web/pnpm-lock.yaml)
- Edited copies of one member (3 members, 18+ corresponding lines) (cmd/generate_changelog/internal/git/walker.go)
- Edited copies of one member (6 members, 15+ corresponding lines) (internal/tools/custom_patterns/custom_patterns.go)
- Edited copy of a member (10 corresponding lines) (internal/plugins/ai/azure/azure.go)
- Edited copy of a member (10 corresponding lines) (internal/plugins/strategy/strategy.go)
- Edited copy of a member (11 corresponding lines) (internal/server/patterns.go)
- Edited copy of a member (13 corresponding lines) (internal/util/groups_items.go)
- Edited copy of a member (14 corresponding lines) (internal/plugins/ai/exolab/exolab.go)
- Edited copy of a member (15 corresponding lines) (cmd/generate_changelog/internal/github/client.go)
- Edited copy of a member (15 corresponding lines) (internal/plugins/ai/copilot/copilot.go)
- Edited copy of a member (16 corresponding lines) (internal/tools/spotify/spotify.go)
- Edited copy of a member (18 corresponding lines) (internal/plugins/ai/gemini/gemini.go)
- Edited copy of a member (18 corresponding lines) (internal/plugins/db/fsdb/patterns.go)
- Edited copy of a member (25 corresponding lines) (internal/domain/attachment.go)
- Edited copy of a member (27 corresponding lines) (cmd/generate_changelog/internal/cache/cache.go)
- Edited copy of a member (30 corresponding lines) (internal/plugins/ai/anthropic/anthropic.go)
- Edited copy of a member (31 corresponding lines) (internal/plugins/ai/azure/azure.go)
- Edited copy of a member (37 corresponding lines) (scripts/python_ui/streamlit.py)
- Edited copy of a member (39 corresponding lines) (internal/plugins/ai/codex/auth_transport.go)
- Edited copy of a member (39 corresponding lines) (internal/plugins/ai/copilot/copilot.go)
- …and 26 more
Changes since last survey
- 93 commits — 83 feature/other, 10 fixes
By area
- cmd/generate_changelog — 40 commits
- (repo) — 30 commits
- internal/plugins — 11 commits
- (root) — 6 commits
- internal/core — 4 commits
- internal/cli — 2 commits
Notable commits
- fix: Merge branch 'main' into fix/yaml-config-merge
- fix: Merge pull request #2320 from jaeminbird/fix-insecure-lan-ipv6-zone-and-entry-normalization
- fix: Merge pull request #2323 from tanishqbaweja/fix/yaml-config-merge
- fix: Merge pull request #2328 from tanishqbaweja/fix/dry-run-session
- fix: chore: fix the ChangeLog snippet
- fix: fix(cli): stop config.yaml from overriding flags and zeroing options it omits
- fix: fix(core): stop --dry-run from applying file changes and saving the session
- fix: fix(plugins): match zoned IPv6 hosts and normalize FABRIC_ALLOW_INSECURE_LAN entries
- fix: fix: localize unconfigured vendor debug messages
- fix: fix: report configured vendor initialization failures
- change: Add changelog for PR #2312: CloudGPU provider
- change: Merge branch 'main' into add-llm-buzz-provider
- change: Merge branch 'main' into add-llm-buzz-provider
- change: Merge branch 'main' into add-wiro-provider
- change: Merge branch 'main' into feat/atmorouter-provider
- change: Merge branch 'main' into feat/cheapaimodel-provider
- change: Merge branch 'main' into feat/runcrate-provider
- change: Merge branch 'main' into feat/standardcompute-provider
- change: Merge branch 'main' into feat/synap-provider
- change: Merge branch 'main' into fix/vendor-configuration-warnings
- …and 73 more
Architecture
- Unchanged — 0 containers · 1 contexts · 0 edges
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
Survey your own repository
danielmiessler/Fabric 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.
About this page
- The score is its most recent published measurement, taken on 13 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
- Measured at commit 793f78f2ec8cb14de84936ba1499b6fc74e5910d — the exact code this score is about.
- Scored under rubric-2026.11.3 — the same rubric and the same method as every other entry in this index.
- Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.61+preprod-b8b1a43ae4c3.