Nightly build

Nightly build from 4d6b79e78bcdea39c4e6d5c78c2ca9c3c2db4cb1.

This prerelease is overwritten on every push to main.

v2.5.0

v2.5.0 - 2026-06-27

Added

• Interactive setup with mods --config, plus first-run configuration when no settings or conversations exist.
• Provider and model setup wizard entries, including web search configuration and clearer base URL handling.
• Continuous chat mode for longer interactive sessions.
• Built-in Mods identity guidance so models understand the CLI, config, tools, and expected behavior.
• Safe workspace filesystem tools (fs_read, fs_write, fs_list, fs_stat) with scoped access.
• Plan mode improvements with multi-proposal selection and local shell command classification.
• Configurable web-search-api-key-env, defaulting to TAVILY_API_KEY.
• Clipboard image shorthand flag.
• Nightly release automation using GoReleaser.

Changed

• Config precedence is now explicitly CLI flags > mods.yml > MODS_ env > defaults.
• Custom OpenAI-compatible endpoints now favor explicit config via mods --config or mods.yml instead of base URL environment auto-discovery.
• Tool approval rules now use directory-scoped grants instead of shell command prefix matching.
• Reasoning controls now actively enable or disable provider-specific thinking behavior, including DeepSeek.
• The build system moved from Mage to Taskfile.
• Streaming, request handling, and tool policy internals were centralized and optimized.
• Thinking and review UI output was softened and simplified.

Fixed

• Respect -T reasoning settings in plan mode and provider-specific reasoning paths.
• Suppress loading animation in debug mode and preserve unsolicited reasoning output.
• Separate model responses cleanly after tool calls.
• Flush cached output and config path updates reliably.
• Validate CLI inputs before provider calls.
• Improve config wizard navigation, provider model setup, base URL screens, and Shift+Tab behavior.
• Fix always-allow directory rule handling and trust LLM shell review analysis.
• Clarify LLM prompt guidance and update README test expectations.

Full comparison: v2.0.0...v2.5.0

v2.0.0

v2.0.0

The first major release of the actively maintained fork. Mods is now an AI
agent for your terminal — not just a pipeline tool. It reads and edits
files, runs shell commands, searches the web, and chains tool calls to finish
a task, with a review step that keeps you in control of anything risky.

⚠️ Breaking Changes

• Module renamed from github.com/charmbracelet/mods to
  github.com/panjie/mods. Update your go.mod and import paths.
• Removed flags: --prompt (-P) and --prompt-args (-p). The -p
  short flag is now used for --plan.
• Default --help trimmed to everyday options; advanced and
  configuration-first options moved to the new --help-all.

✨ Highlights

• Plan mode (--plan / -p): generate a detailed plan for your approval
  before Mods touches any files or runs any commands.
• Arrow-key navigation in approval prompts — accept, deny, or save a
  per-conversation rule without leaving the home row.
• Multi-provider reasoning (--reasoning on|off|auto, -T): unified
  reasoning config works across OpenAI, Anthropic, DeepSeek, GLM, Qwen, Kimi,
  and MiniMax, with distinct rendering of the model's thinking.

🔌 New Providers

Added four OpenAI-compatible providers with model templates and reasoning
support:

• GLM — glm-5.2
• Qwen — qwq-plus, qwen3-plus
• Kimi — kimi-k2
• MiniMax — MiniMax-M3

Model list refreshed across OpenAI, Anthropic, Google, Cohere, DeepSeek,
OpenRouter, and Ollama.

🔒 Safety & Stability

• Closed critical security gaps in fs_apply_patch, Google auth, and web
  search input handling.
• Hardened tool/shell execution safety and concurrency.
• Fixed cache atomic writes, provider stream contracts, and flag parsing.
• Plan output is now preserved across the approval flow.

🛠️ CLI & DX

• --help-all groups every flag by purpose (Model, Session, I/O, MCP,
  Tools, Debug, …).
• Expanded CLI examples covering image, web search, plan, review, MCP, and
  reasoning.
• Release archives expanded to Linux and macOS (previously Windows-only).

📚 Documentation

• README rewritten to showcase Mods' agent capabilities, with high-impact
  examples for code review, refactoring, debugging, and scaffolding.
• Removed stale standalone docs and the legacy fanciness option.

Full Changelog: v1.0.1...v2.0.0

v1.0.1

Changelog

• 502c778 Prepare Windows release archives for WinGet
• 9990c6f Remove manpage support and use XDG install dirs
• 51dc753 build: replace Makefile with Mage
• af971d8 build: add install targets with XDG support
• d56faef refactor: move CLI internals under internal packages
• 11b2c6d add demo.tape
• a98a100 add demo.gif
• 0fc6381 Update image links in README to assets directory
• 54fa9ee chore: update .gitignore
• 9e6caa1 Add assets directory with demo.gif and mods-product.png
• 5d994be chore: remove examples dir from git tracking
• fd6e35e Add make target for manpage generation
• ddb974c Keep generating animation until output starts
• a9b4ee4 Add option to hide tool status
• 0e0e7c5 chore: ignore .DS_Store and .gif files
• 73a2180 Update example tape gif generation
• 8feb6b4 Use system mods for example tapes
• 6e4b1ef Refresh local README example GIFs
• a557c46 docs: add demo tape examples
• 08d1e9b docs: align README with current project state
• e964154 docs: fix badges, license, completions text, and typo
• 55d34c3 docs: replace package manager install with build from source
• 275015e fix: add missing config entries to template and help map, fix --theme default
• aab4205 feat: change newline key binding to ctrl+j in prompt editor
• 36c4956 fix: harden shell approval, fix provider stream contracts, simplify shell classifier to LLM-only
• ed423a1 fix: harden tool/shell execution safety and concurrency, refactor approval chain
• 7861e5c Refactor shell tool execution safety
• c694971 Add direct PowerShell tool on Windows
• caa1182 Fail closed when tool review is unavailable
• 318d0c9 Make Makefile targets Windows-compatible
• 02bec8d feat: configurable workspace root and shell UX improvements
• 8aa7660 feat: detect actual shell for approval rules, support non-POSIX shells
• 535c73d improve: shell command classification, word wrap handling, and tool selection hints
• 4abd9a0 add conversation-scoped approval rules
• d6539da fix: tell LLM which shell and path format to use
• c5e26e9 fix: treat non-zero shell exit codes as successful tool calls
• 307a7ac fix: decode cmd.exe output from system ACP to UTF-8 on Windows
• fa0d768 fix: prevent nil pointer panic in Google stream when API request fails
• b93541f refactor: extract shared provider config factory, image helper, and tool round limit logic
• f648f96 fix: handle glamour init error, fix mutex copy, harden config perms, guard nil db
• e676584 chore: set version to 1.0.0, add CHANGELOG, inject version in dev build
• e8085a6 fix: use markdown hard line breaks (trailing spaces) instead of paragraph breaks to preserve single newlines
• 86b7222 fix: preserve line breaks in glamour output by converting single newlines to double
• 5f6ad82 fix: remove MaxTokens=10 limit in shell classification to prevent empty LLM responses
• b9fe096 fix: debug flag not syncing to atomic bool; improve shell classifier reliability
• 74109b4 fix: complete Azure OpenAI (non-AD) support end-to-end
• 4a1ed1e chore: remove unused max-completion-tokens field
• 5b11f50 feat: LLM-based shell command classification for mutable review mode
• 44a1dad refactor: remove shell command safety heuristics from review system
• 2832425 Sort --help options alphabetically by removing SortFlags=false
• c305f7f fix: address bugs found in code review
• ecf9eb4 refactor: split Config, extract client factory, split mods.go
• acc3b98 refactor: extract tool reviewer, remove global config, fix naming
• e7673a4 fix: address concurrency bugs, error handling gaps, and code duplication
• 55f1e0b fix: skip review in non-interactive mode to prevent deadlock
• d23fe0b docs: add review banner visual and config YAML example to README
• a95304a docs: update README for actively maintained fork
• 98ac581 fix: close reviewChan, handle ctrl+c deadlock, remove stash from harmless git cmds
• 0bafbd9 feat: add tool execution review mechanism with configurable shell heuristics
• 35231b7 feat: add unified --reasoning option with auto mode and thinking debug output
• 16d08f8 feat: add release target to Makefile
• 67ffd15 build: standardize local build output
• 11bec70 fix: cross-platform compatibility fixes
• f26585a feat: show status line while model generates between tool calls
• 590db93 Add minimal pipeline output mode
• a7efbcc Use DuckDuckGo for default web search
• f4f6d17 Improve tool operation status output
• d179105 refactor: remove web_search pre-fetch logic, unify with other built-in tools
• 31f9ba2 feat: update model list to latest releases (H2 2025 - H1 2026)
• 5e7742e refactor: streamline config_template.yml — trim to essential providers
• e3443ee feat: change default web search provider from Bing to Google
• 1bd17e0 feat: add --max-tool-rounds with failed/success dual-track limiting
• c16c646 fix: update tests for refactored tool system (fromMCPTools → fromToolSpecs, Tool removed)
• 25ba52b fix: code review improvements — bug fixes, test coverage, and hardening
• 4075e5e Gate filesystem tools automatically
• 1a6864a Add native tool registry
• 0e51531 fix mcp cancellation and config paths
• a9489a1 fix: cross-platform HideWindow via platform-specific helpers
• 75700de fix(windows): suppress console popup and reset-settings file lock
• c75565e feat: add tool result truncation, round limit, MCP whitelist, and tests
• 37e24d8 fix: prevent 413/token overflow with tool result truncation, system context, and round limiting
• 7c87c50 chore: update API model configurations for DeepSeek, Cohere, OpenAI, Anthropic, Google
• f1c1030 feat: add --debug/-D flag for execution step and request diagnostics
• 18c294d fix: strip MCP tool schema descriptions and handle HTTP 413/4xx errors
• 4a6df7c feat: add clipboard image support with cross-platform compatibility
• 0e1c3c3 feat: add image recognition support for vision-capable LLMs
• b97a82c docs(README): document web search and show-tool-calls features
• 4f7a3b6 feat: add show-tool-calls config option to control tool call display
• 1c78c7d fix: remove hardcoded setlang=en in Bing search
• 656bcd8 Merge pull request #1 from panjie/feature/web-search
• a7011c3 feat: add web search capability (Bing/Tavily)
• 0425d0d docs(README): update sunset notice
• cb1bf54 chore(deps): bump the all group with 3 updates (charmbracelet#677)
• d6bbd4e chore(deps): bump the all group with 2 updates (charmbracelet#676)
• 364aae5 chore(deps): bump the all group with 3 updates (charmbracelet#673)
• 941bf5a chore(deps): bump the all group with 5 updates (charmbracelet#672)
• f6747c8 chore: update LICENSE copyright
• 0a17796 docs(readme): add sunset notice (charmbracelet#668)
• e3cc766 chore(deps): bump the all group with 3 updates (charmbracelet#667)
• b47ca22 chore(deps): bump the all group with 4 updates (charmbracelet#665)
• 1e22839...