Releases
v2.0.0
Compare
Sorry, something went wrong.
No results found
What's changed
✂️ Updated Features
Reframed from an OpenAI-Assistant shell into a provider-agnostic library of safe,
token-aware filesystem tools for any LLM agent.
Unified diffs are now the primary edit path: apply_git_patch validates that
every target file is inside the root folder, and (when auto_cat is on) returns
the patched file contents so a clean exit isn't mistaken for a correct edit.
Tools are now invoked through a neutral ToolKit.dispatch(name, arguments)
instead of an OpenAI-run object.
The TODO task store now lives at ai_shell.todo and is stored as Markdown with
YAML frontmatter (TODO.md / TODO_done.md) instead of a private TOML format.
Tasks gained a done_when acceptance-criteria field for the goal loop.
Read-only git tools now shell out to the git CLI; gitpython is no longer a
direct dependency.
Linters/formatters/test-runners are now an optional checkers extra rather than
required dependencies.
Packaging migrated from Poetry to standard PEP 621 (uv).
🚫 Removed
The OpenAI Assistants (Beta) bot runtime: TaskBot, the tool loop, the demo
bots, the markdown dialog logger, and the openai / openai_multi_tool_use_parallel_patch
dependencies. Bring your own agent loop.
The ed and edlin line-editor tools — edit via unified diffs (apply_git_patch)
or replace/insert/rewrite instead.
Unimplemented stubs (regex-tester, tool-picker sub-bot) and their config flags.
The OpenAI-API JSON request logger and bundled sample logs / demo fixtures.
bug-trail-core, markpickle, and doc-only markdown dependencies from the core
install.
You can’t perform that action at this time.