1.0.3

v1.0.3

Bug fixes and quality-of-life improvements. Same install command:

pip install --upgrade agent-code

1.0.2

v1.0.2

Bug fixes and quality-of-life improvements. Same install command:

pip install --upgrade agent-code

1.0.1

v1.0.1

Bug fixes and quality-of-life improvements. Same install command:

pip install --upgrade agent-code

Added

• Markdown rendering in the chat log — headers, bold, italic, inline
  code, fenced code blocks, bullets, numbered lists, blockquotes, links,
  and horizontal rules all render properly instead of showing raw ** and
  backticks.
• Clear button — wipes the log and conversation memory (with confirm).
• Copy last button — copies the last agent answer to the clipboard.
• Export button — saves the full chat as a formatted markdown file.
• Todo button in the top toolbar — shows the persistent task list.
• Permission modes — readonly, ask, auto. Controls whether the
  agent can edit files and run shell commands without confirmation.
• Subagent tool (task) — spawns an isolated agent with a fresh
  context for research and exploration, so the main conversation stays
  clean.
• Todo list (todo_write / todo_read) — the agent can plan multi-step
  work and update progress across turns.
• web_fetch(url) tool — one-shot URL fetch that strips HTML, returns
  readable text. Faster than spinning up the browser for simple pages.
• edit_file tool — exact-string replacement for targeted edits.
  Preferred over write_file because it doesn't require reproducing the
  whole file.
• glob_files and grep_files — find files by pattern, search contents
  with regex.
• read_file with offset and limit — line-numbered output, safe on
  large files.
• Project context loading — auto-reads AGENT.md, CLAUDE.md,
  AGENTS.md, or README.md from the working directory into the system
  prompt.

Changed

• Windows config location — everything now lives under
  %LOCALAPPDATA%\agent-code\ instead of scattered across your user
  profile:
  • %LOCALAPPDATA%\agent-code\config\ — agent_config.json,
    skill-manager.json
  • %LOCALAPPDATA%\agent-code\data\
  • %LOCALAPPDATA%\agent-code\cache\
  • %LOCALAPPDATA%\agent-code\state\
  • ~/.config/agent-code/ on Linux/macOS (unchanged)
• Polished UI — deeper background, layered panels, framed log card,
  fixed-height header and footer bars, three button styles (Icon / Stop /
  Send), blue input focus ring, better spacing throughout.
• Dual-mode agent.py — the same file works both as
  python man-code\agent.py (standalone) and as agent-code (installed
  package). No duplicate maintenance.
• System prompt rewritten — Claude Code-inspired workflow: plan with
  todo_write, read before editing, prefer edit_file over write_file,
  verify after edits, use subagents for exploration.
• Turn limit raised from 30 to 40 for complex multi-step tasks.

Fixed

• agent.py no longer uses the old keywords-based scanner — the server
  does the AI review, and the client just shows the verdict.
• NameError: cannot access free variable 'e' in Tkinter callbacks
  when a network error occurred. Messages are now captured into a local
  variable before being passed to the lambda.
• AttributeError: 'tkapp' object has no attribute 'flagged_only' in
  the skill store — all UI state flags are now initialized in __init__.
• Path resolution on NixOS, containers, and headless environments —
  paths.py now handles missing $HOME, read-only root filesystems, and
  falls back gracefully to CWD or /tmp.
• Cloud sync — the /state endpoints now correctly handle all response
  shapes, and server fallback works reliably when the primary HTTPS domain
  is down.
• Browser self-test — runs on startup and reports ready or FAILED
  in the header.

Not changed

• License: MIT (client) + GPL-3.0 (server)
• Python requirement: 3.10+
• Bundled skills: filesystem, public_api, temp_mail
• Registry server: https://skills-manager.freesrv.com
• Fallback IP: http://78.154.103.43:9074

Install

pip install --upgrade agent-code

Try

agent-code                  # run the agent
agent-code manage           # open the skill store
agent-code skills install   # copy bundled skills to ./skills/
agent-code --help           # all commands

1.0.0

v1.0.0 — Initial Release

First public release of Agent + Skill Store — a local AI agent with a
headless browser, sandboxed shell, and a plug-and-play skill system.

What's included

Agent (agent.py)

• Headless Chromium browser (DrissionPage) — no popup window
• Sandboxed shell — allowlist + workspace/ jail
• Dark tkinter GUI with chat log and input
• Stop / Continue buttons — halt mid-turn, resume from last tool result
• Local chat auto-save (chats/current.json)
• Optional cloud sync with HTTPS → IP fallback
• Multi-key LLM rotation + custom base_url
• Settings, Skills, and two-pane Help dialogs

Skill manager (skill-manager.py)

• GUI + CLI for the skill registry
• Sign in, browse, install, publish skills
• Server fallback: https://skills-manager.freesrv.com → http://78.154.103.43:9074

Bundled skills

• filesystem — read/write/append/list files in the sandbox
• public_api — search 700+ free public APIs and call them
• temp_mail — temp inbox + email sending (Resend / Brevo / Mailjet)

Docs

• README.md — full setup and usage guide
• docs/skills.html — developer guide for writing skills

Install

pip install DrissionPage openai
git clone https://github.com/minecraftbefile-maker/Agent-code.git
cd Agent-code
cp agent_config.example.json agent_config.json
# edit agent_config.json with your API key
python agent.py

Windows: double-click run.bat instead.

Requirements

• Python 3.10+
• Windows, Linux, or macOS
• An OpenAI-compatible API endpoint (OpenAI, Together, Groq, DeepSeek, local vLLM, …)

License

• Client code (agent.py, skill-manager.py, skills/**, docs/**): MIT
• Registry server (skill-server.py): GPL-3.0

Not included

skill-server.py is not in this repo. A live instance is hosted for you at
https://skills-manager.freesrv.com — both clients use it by default.
If you want your own, the source is available separately under GPL-3.0.