Changelog

Full release history. Same content as m changelog in the CLI.

v0.0.33
May 2026
  • Bundled agents — all 32 agents embedded in the binary, no clone needed
  • Instant access — m chat devops works right after install
  • On-demand extraction — agents extracted to config dir on first use
  • User edits preserved — extracted agents never overwritten
  • Windows CI fixes — bash shell for gosec, JSON path escaping, .exe suffix
v0.0.32
May 2026
  • Windows support — shell uses cmd.exe, Windows Credential Manager, .exe in releases
  • Structured logging — slog JSON output, debug/warn levels
  • Graceful shutdown — session saved on Ctrl+C / SIGTERM
  • Integration tests — 10 black-box CLI tests, CI on Windows + Linux
  • TUI fix — ANSI-aware word wrap
v0.0.31
May 2026
  • PII guardrails — redact emails, phones, SSNs, credit cards, API keys before sending to LLM
  • Mermaid diagram skill — teaches agents to produce architecture diagrams
  • SAST + SCA — gosec and govulncheck in CI pipeline
  • Secret scanning — push protection enabled
v0.0.30
May 2026
  • Update notifier — checks GitHub once/day, shows notice if newer version available
v0.0.29
May 2026
  • REPL color — bold blue prompt, dim tips
  • Help examples — real usage in m chat/run --help
  • Session summary — /sessions shows first user message
  • Tool output — line count instead of raw bytes
v0.0.28
May 2026
  • Command shortcuts — /x /r /c /u /m /t /s /h for all slash commands
  • TUI command bar — underlined shortcut letters
  • Welcome shortcuts — listed on every session start
v0.0.27
May 2026
  • First-run wizard — visual completion summary with next steps
  • Onboarding tips — key commands shown on every session start (TUI + REPL)
v0.0.26
May 2026
  • m new — scaffold a new agent .md file with boilerplate
  • m doctor — health check: config, API key, model reachability, tools
  • m completion — shell completions for bash/zsh/fish/powershell
  • Named sessions — /save fixing-auth-bug creates a named snapshot
  • Better errors — API key errors now tell you how to fix them
v0.0.25
May 2026
  • MCP HTTP/SSE transport — all three transports now supported: stdio, HTTP, SSE
  • Agent registry — m install + run/chat by name without path
  • Shell timeout — 30s → 120s for terraform/docker
  • Undo stack capped at 20 entries
  • Example MCP servers — datadog (HTTP), slack (SSE)
v0.0.24
May 2026
  • code_search tool — codebase RAG: grep + in-memory symbol index (functions, types, classes, imports)
  • 9 languages indexed — Go, Python, JS, TS, Java, Ruby, Rust, Terraform, Shell
  • Session history rotation — autosave keeps last 10 backups
  • /trust command — auto-approve tools in TUI (dangerous commands still double-confirm)
  • Dangerous command protection — 34 patterns always require double y/n
v0.0.23
May 2026
  • Fallback models — auto-switch on 429 rate limit. All 30 example agents updated
  • Per-agent thinking phrases — customize spinner text (e.g. Serbian for Steva Đubre)
  • Markdown rendering — bold, code, headers rendered as terminal styles
  • Tool confirmation — clean key=value format instead of raw JSON
  • Tool output formatting — proper line breaks between model text and tool activity
  • TUI banner — version + copyright + tagline
v0.0.22
May 2026
  • Error intervention — when a tool fails, the REPL lets you type a hint that gets sent back to the LLM, breaking retry loops
  • Anthropic prompt caching — system prompt and tool schemas use cache_control: ephemeral, cutting input tokens on every turn
  • Dead code cleanup — removed unused stdinToolConfirm / isSafeTool, fixed pre-existing compile issues in chat state init
v0.0.21
July 2026
  • High-contrast confirmations — ConfirmFg theme field, all 9 themes updated. y/n prompts are now always readable regardless of terminal background
  • TUI throttle bypass — important messages (tool output, confirmations) skip the 60fps throttle for instant display
  • Lint fixes — gocritic if-else chain rewritten to switch, go fmt formatting
  • CI validation — validate all examples/ so MCP/skill cross-references resolve
  • First self-made release by Steva Đubre (the agent fixed himself!)
v0.0.19
July 2026
  • Homebrew install — brew tap subzone/tap && brew install subzone/tap/m
  • Auto-update formula on every release via CI
  • Open source community files — SECURITY.md, CODE_OF_CONDUCT.md, CODEOWNERS, issue/PR templates
  • CI on every push/PR — vet, lint, race tests, build, validate agents
  • Dependabot for Go modules and GitHub Actions
  • Branch & tag protection, repo topics for discoverability
v0.0.18
July 2026
  • Session persistence — autosave after every step, /save /sessions /resume. AES-256-GCM encrypted, key in OS keychain
  • web_fetch tool — fetch URLs and extract readable text. Stdlib-only, no API key
  • /models command — numbered model picker with live API scanning. Token plan probing for DashScope
  • 9 built-in themes — nord, dracula, gruvbox, tokyonight, catppuccin, solarized + original 3
  • m list — agent discovery command, scans directories for .md agents
  • Token-based context compaction — per-model context windows (80% budget), meaningful summaries of dropped messages
  • Reasoning model support — MiniMax-M2.5, DeepSeek-R1 thinking indicator
  • Alibaba token plan — custom base URL, GLM-5, MiniMax-M2.5 via DashScope
  • TUI guardrails — auto-approve read-only tools, y/n for destructive (shell, fs_write, git)
  • TUI scrolling fix — eliminated jitter during streaming
  • DashScope filtering — 154 → 45 models (skip non-chat, deduplicate dated snapshots)
  • Screenshot gallery on landing page
v0.0.17
May 2026
  • Product landing page — professional site at subzone.github.io/Agentctl, pure HTML/CSS, Font Awesome icons, no Jekyll
  • 7-page documentation site — home, install (platform tabs), agents gallery, architecture (comparison vs Cursor/Copilot/Aider), MCP servers, changelog timeline, support
  • DevOps agents — k8s-debug (Kubernetes triage), terraform-plan (plan review), helm-deploy (chart management)
  • Jira/Confluence integration — ticket-worker (ticket → branch → code → test → update), ticket-reviewer (review against acceptance criteria)
  • MCP server definitions for Jira and Confluence (mcp-server-atlassian)
  • Steva Đubre & Steve Trash personality agents — Serbian/English, multiple model variants
  • ASCII art M banner in site nav and footer
v0.0.12
May 2026
  • Theme system — built-in themes (matrix, default, minimal) + custom themes via ~/.config/m/theme.yaml
  • Matrix theme is the new default: green monochrome
  • /theme command — list themes or switch live
  • Responsive layout — header collapses on small terminals
  • Tool activity visible in TUI — → fs_list / ← 245 bytes
  • Visual message styling — user (bold), tools (faint), errors (red)
  • Gemini/Alibaba/LiteLLM fix — WithCompat() disables OpenAI-specific stream_options
v0.0.11
May 2026
  • Google Gemini provider — gemini-2.5-pro/flash via OpenAI-compat
  • Alibaba Cloud provider — qwen-plus/turbo/max via DashScope
  • m config command — interactive provider/model manager with model discovery
  • Wizard updated to 6 providers
  • Full provider/model label below the token box in TUI
v0.0.10
May 2026
  • Structured output post-processing — engine buffers JSON, extracts answer field
  • Anthropic response-tool extraction — synthetic tool call intercepted
  • Parallel tool execution — multiple tool calls run concurrently
  • Ports/adapters layer — ConfigSource, Secrets, StateStore interfaces + MemoryStore
v0.0.9
May 2026
  • Engine-enforced structured output — response_schema in frontmatter constrains model to valid JSON
  • Hub-and-spoke agents — orchestrator + spoke-coder/reviewer/planner with citations
  • 429 retry with backoff for Anthropic and OpenAI
  • Ollama output cap — num_predict defaults to 8192
v0.0.8
May 2026
  • /model command — switch LLM mid-session, history preserved
  • /compact — truncate history to last 4 exchanges
  • Context window indicator — ctx: N% on input line
  • Auto-continue on max_tokens — engine sends "continue" automatically
  • fs_write + fs_list tools — create/patch files, list directories
  • Orchestrator agent — routes to 6 specialists
v0.0.6 – v0.0.7
May 2026
  • Token count + cost in TUI header
  • fs_write tool — create/patch with user confirmation + undo
  • fs_list tool — recursive, skips .git/node_modules
  • New agents — reviewer, writer, devops, local
  • Makefile + golangci-lint + CI gate
v0.0.3 – v0.0.5
May 2026
  • Chat TUI — full-screen bubbletea UI with banner, stats, scrolling chat
  • Auto-fallback to line REPL when not a TTY
  • thinking… spinner while waiting for first token
  • Crash fix — strings.Builder copied by value panic
  • Ollama daemon detection — 15s poll with backoff
v0.0.1
May 2026
  • Initial release
  • Default chat — type m to talk
  • Setup wizard — Ollama, Anthropic, OpenAI, LiteLLM
  • Keychain-backed key storage
  • Tag-driven release pipeline — macOS .pkg + Linux .deb