conferllm 0.1.0__tar.gz

conferllm-0.1.0/.gitignore

@@ -0,0 +1,137 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+.DS_Store
+.ruff_cache
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+# uv keeps the real environment in its cache. Ignore the compatibility link
+# and any numbered copies created by synced-folder conflict resolution.
+/.venv*
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+.streamlit/
+coverage.xml
+htmlcov
+.claude
+.mypy_cache
+.ruff_cache 
+.pytest_cache

conferllm-0.1.0/AGENTS.md

@@ -0,0 +1,48 @@
+# Repository Guidelines
+
+## Project Structure & Module Organization
+- Source: `src/conferllm/` (`cli.py`, `chat.py`, `session.py`, `artifacts.py`, `images.py`, `server.py`, `client.py`, `config.py`, `doctor.py`, `skill.py`)
+- Tests: `tests/` with shared fixtures in `tests/conftest.py`
+- Config template: `config_example.yaml` (copy to `~/.conferllm/config.yaml`)
+- Skill bundle: `skills/conferllm/` (`SKILL.md`, `agents/openai.yaml`, `references/`)
+- Packaging & tools: `pyproject.toml` (ruff, mypy, pytest), console script `conferllm`
+- Docs: `README.md`, `CLAUDE.md`, `AGENTS.md`
+
+## Build, Test, and Development Commands
+- Setup: use `uv>=0.12.9,<0.13`, then run `uv sync --extra dev`
+- Run server: `uv run conferllm serve [--transport stdio|sse|http] [--config PATH] [--log-level DEBUG]`
+- New chat: `uv run conferllm chat --model MODEL --prompt TEXT`
+- Continue chat: `uv run conferllm chat --session SESSION_ID --prompt TEXT`
+- List sessions: `uv run conferllm sessions list`
+- Tests: `uv run pytest`
+- Coverage: `uv run pytest --cov=src/conferllm --cov-report=term-missing`
+- Lint/format: `uv run ruff format . && uv run ruff check .`
+- Type check: `uv run mypy src/`
+- Build: `uv build`
+
+## Coding Style & Naming Conventions
+- Python 3.10+, PEP 8, 4-space indent, line length 88 (ruff configured)
+- Type-annotate all functions; keep imports sorted (ruff `I`); avoid unused args
+- Files/modules use `snake_case`; classes `PascalCase`; constants `UPPER_SNAKE_CASE`
+
+## Testing Guidelines
+- Frameworks: pytest, pytest-asyncio, pytest-cov
+- Discovery: files `test_*.py` or `*_test.py`; classes `Test*`; functions `test_*`
+- Add tests for new behavior and failure paths; use mocks (no real API calls)
+- Maintain or increase coverage; HTML report emitted to `.cache/coverage/html/`
+
+## Commit & Pull Request Guidelines
+- Use Conventional Commits (e.g., `feat:`, `fix:`, `docs:`, `chore:`); imperative, present tense
+- Keep commits focused; exclude generated artifacts (`htmlcov/`, `.coverage`, `coverage.xml`)
+- PRs include: clear summary and rationale, linked issues, reproduction steps or command output, and test plan/results
+
+## Security & Configuration Tips
+- Never commit secrets; put real keys only in `~/.conferllm/config.yaml`
+- Keep `~/.conferllm` at mode `0700` and `~/.conferllm/config.yaml` at mode `0600`
+- Session JSONL may contain sensitive conversations; tests and tools must not read real files under `~/.conferllm/sessions/`
+- Tests and docs must not read or print real local configuration values
+- Prefer LiteLM provider/model IDs (e.g., `openai/gpt-4`, `anthropic/claude-3-5-sonnet-20241022`)
+
+## Agent-Specific Instructions
+- Treat this file as authoritative across the repo scope
+- Keep changes minimal and idiomatic; run `ruff`, `mypy`, and `pytest` before large refactors

conferllm-0.1.0/CLAUDE.md

@@ -0,0 +1,115 @@
+# CLAUDE.md
+
+This file provides project guidance for Claude Code and compatible coding
+agents.
+
+## Project overview
+
+ConferLLM exposes locally configured LiteLLM models through:
+
+- the stateful `conferllm chat` CLI;
+- an MCP server over stdio, SSE, or streamable HTTP;
+- the `conferllm` Agent Skill in `skills/conferllm/`.
+
+The default configuration path is `~/.conferllm/config.yaml`. Never read, print,
+commit, or otherwise expose real credentials from that file. Session JSONL is
+stored below `~/.conferllm/sessions/` and may also contain sensitive content.
+
+## Architecture
+
+```text
+src/conferllm/
+├── __init__.py   # Package version and public exports
+├── __main__.py   # python -m conferllm
+├── cli.py        # CLI and MCP-server command routing
+├── chat.py       # Shared stateful chat orchestration
+├── client.py     # LiteLLM wrapper and local image conversion
+├── config.py     # Pydantic YAML configuration models
+├── doctor.py     # Safe installation/configuration diagnostics
+├── errors.py     # Typed, stable public error mapping
+├── artifacts.py  # Transactional session-owned image artifacts
+├── images.py     # Input validation and output-image extraction
+├── session.py    # JSONL session persistence and filtering
+├── server.py     # MCP tools, resources, and transports
+└── skill.py      # Version-matched bundled Skill installer
+```
+
+The main flows are:
+
+```text
+Agent or user -> conferllm CLI -> ChatService -> SessionStore + LLMClient
+MCP client    -> MCPServer   -> ChatService -> SessionStore + LLMClient
+```
+
+Internal adapters use the concrete `ChatResult` and
+`SessionListResult` interfaces; do not add runtime reflection or permissive
+fallback branches to accommodate test doubles.
+
+YAML model aliases are unique, provider model IDs are required, and relative
+session roots resolve beside the configuration file. Configuration and provider
+exceptions must be sanitized at their source; public codes must not be inferred
+from message wording.
+
+For continuation, keep the lock across load, capability/history checks, orphan
+recovery, provider execution, and JSONL commit. JSONL is the commit record.
+Never discard a committed turn, silently truncate history, or retry the model
+because an optional export/inline-image read failed after commit.
+
+## Development commands
+
+```bash
+uv sync --extra dev
+uv run conferllm --help
+uv run pytest
+uv run ruff format .
+uv run ruff check .
+uv run mypy src/
+uv build
+```
+
+Run dependency discovery without changing `pyproject.toml`:
+
+```bash
+uv run scripts/upgrade_dependency.py --dry-run
+```
+
+## CLI contract
+
+- `conferllm` displays top-level help.
+- `conferllm serve` is the only MCP server entry point.
+- `conferllm chat --model MODEL --prompt TEXT` creates a named session.
+- `conferllm chat --session SESSION_ID --prompt TEXT` continues stored history.
+- `conferllm sessions list` lists and filters session metadata.
+- `conferllm models` lists configured aliases.
+- `conferllm model-info MODEL` returns non-secret model metadata.
+- Legacy root-level server options are not supported.
+
+Do not change these interfaces without updating README, tests,
+`skills/conferllm/SKILL.md`, and `skills/conferllm/agents/openai.yaml` together.
+
+## MCP tools
+
+- `create_chat(message, model, name=None, images=None, image_output_dir=None, include_raw_response=False)`
+- `continue_chat(message, session_id, images=None, image_output_dir=None, include_raw_response=False)`
+- `chat(message, model=None, session_id=None, name=None, images=None, image_output_dir=None, image_path=None, include_raw_response=False)` (compatibility wrapper)
+- `list_sessions(query=None, model=None, since=None, until=None, limit=50)`
+- `list_models()`
+- `get_model_info(model)`
+
+The code uses MCP 2.x `MCPServer`. Direct `call_tool()` tests receive a
+`CallToolResult`; use its `content` and `structured_content` fields. Generated
+images may be returned inline as `ImageContent` or by
+`conferllm://sessions/{session_id}/artifacts/{artifact_id}` resource URI.
+The first text block contains the same JSON envelope as structured content so
+text-only clients can retain the session ID. Keep expensive listing, replay,
+and inline-artifact I/O off the event loop.
+
+## Testing
+
+All provider calls must be mocked. Add tests for success and failure paths,
+including CLI exit codes and transport mapping. Keep or improve coverage.
+`tests/conftest.py` isolates `Path.home()` and rejects unmocked provider calls.
+Do not disable these guards for a smoke test or inspect the real user's files.
+
+Before handing off a substantial change, run formatting, lint, type checking,
+tests, `git diff --check`, package build, and Skill validation.

conferllm-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Pengfei Ni
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

conferllm-0.1.0/PKG-INFO

@@ -0,0 +1,378 @@
+Metadata-Version: 2.5
+Name: conferllm
+Version: 0.1.0
+Summary: A local CLI and MCP server for talking to configured AI models through LiteLLM.
+Project-URL: Homepage, https://github.com/feiskyer/mcp-ai-hub
+Project-URL: Repository, https://github.com/feiskyer/mcp-ai-hub
+Project-URL: Documentation, https://github.com/feiskyer/mcp-ai-hub#readme
+Project-URL: Bug Tracker, https://github.com/feiskyer/mcp-ai-hub/issues
+Author-email: Pengfei Ni <feiskyer@gmail.com>
+License: MIT License
+        
+        Copyright (c) 2025 Pengfei Ni
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Requires-Python: >=3.10
+Requires-Dist: httpx[socks]>=0.28.1
+Requires-Dist: litellm<1.98.0,>=1.97.0; python_version < '3.11'
+Requires-Dist: litellm<2,>=1.99.0; python_version >= '3.11'
+Requires-Dist: mcp<3,>=2.1.1
+Requires-Dist: pydantic<3,>=2.13.5
+Requires-Dist: pyyaml>=6.0.3
+Provides-Extra: dev
+Requires-Dist: mypy>=2.3.1; extra == 'dev'
+Requires-Dist: pre-commit>=4.6.2; extra == 'dev'
+Requires-Dist: pytest-asyncio>=1.4.0; extra == 'dev'
+Requires-Dist: pytest-cov>=7.1.0; extra == 'dev'
+Requires-Dist: pytest>=9.1.1; extra == 'dev'
+Requires-Dist: ruff>=0.16.6; extra == 'dev'
+Requires-Dist: types-pyyaml>=6.0.12.20260815; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# ConferLLM
+
+ConferLLM is a local command-line tool for asking configured AI models questions,
+continuing conversations, and working with images. Its bundled Agent Skill lets
+another agent use the same commands for a second opinion or a model comparison.
+
+The Python package, import package, executable, and Skill are all named
+`conferllm`. LiteLLM handles provider integration; callers select a local model
+alias and keep the returned session ID.
+
+## Install
+
+From an authorized ConferLLM source checkout:
+
+```bash
+uv tool install .
+```
+
+If you already use a Python virtual environment:
+
+```bash
+python -m pip install .
+```
+
+A supplied ConferLLM wheel can be used instead of `.`. Once a release of this
+project has been published and its identity verified, it can be installed with
+`uv tool install conferllm`. Do not assume that a same-named package is this project.
+
+Verify the executable:
+
+```bash
+conferllm --version
+conferllm --help
+```
+
+ConferLLM requires Python 3.10+ and POSIX file locking (macOS/Linux). Native Windows
+is not supported.
+
+## Configure models
+
+Configuration belongs in `~/.conferllm/config.yaml`, outside the repository.
+From a source checkout, create a private directory and copy the example only
+when no configuration exists:
+
+```bash
+mkdir -p ~/.conferllm
+chmod 700 ~/.conferllm
+test ! -e ~/.conferllm/config.yaml && cp config_example.yaml ~/.conferllm/config.yaml
+chmod 600 ~/.conferllm/config.yaml
+```
+
+Alternatively, create the file using this minimal configuration and supply
+your own credentials:
+
+```yaml
+model_list:
+  - model_name: reasoning
+    capabilities:
+      input_modalities: [text]
+      output_modalities: [text]
+    litellm_params:
+      model: openai/gpt-5
+      api_key: "replace-with-your-key"
+
+  - model_name: vision
+    capabilities:
+      input_modalities: [text, image]
+      output_modalities: [text]
+      max_input_images: 8
+    litellm_params:
+      model: openai/gpt-4o
+      api_key: "replace-with-your-key"
+```
+
+Provider model IDs and parameters are examples; availability depends on your
+provider account. Each local alias must be non-empty and unique, and
+`litellm_params.model` must identify a provider model. See
+[config_example.yaml](config_example.yaml) for additional configuration shapes.
+
+Values in `litellm_params` are forwarded to LiteLLM, except that ConferLLM owns
+`messages` and always sets `stream` to `false`. Unknown application configuration
+fields are rejected so typos do not silently change behavior.
+
+Optional settings:
+
+- `global_system_prompt`: applied to every request.
+- A model's `system_prompt`: overrides the global prompt; `""` disables it.
+- `sessions_dir`: overrides the session root. Relative YAML paths resolve
+  beside the configuration file, independently of the caller's working directory.
+- `image_limits`: limits newly attached images per turn. Defaults are 16 images,
+  20 MiB per image, and 50 MiB in total.
+
+Use `--config PATH` to select another file. An explicitly selected missing file
+is an error. ConferLLM does not create, populate, or automatically migrate provider
+credentials.
+
+Check configuration without printing its values:
+
+```bash
+conferllm doctor --json
+conferllm models --json
+conferllm model-info reasoning
+```
+
+Doctor reports safe metadata and suggested next steps. It exits nonzero when
+configuration is missing, invalid, or has no model aliases, while still
+returning its diagnostic report.
+
+## Ask a model
+
+Create a conversation:
+
+```bash
+conferllm chat --model reasoning --prompt "Explain Raft leader election."
+```
+
+Human-readable output includes the session ID, name, answer, output-image
+locations, and any warnings. A name is derived locally from the first prompt;
+use `--name` to choose one explicitly.
+
+For scripts and agents, request structured output:
+
+```bash
+conferllm chat \
+  --model reasoning \
+  --name "Raft notes" \
+  --prompt "Explain Raft leader election." \
+  --json
+```
+
+The stable response envelope is `conferllm.chat.response.v1`. Its authoritative
+fields are:
+
+```json
+{
+  "schema_version": "conferllm.chat.response.v1",
+  "ok": true,
+  "session": {
+    "id": "20260905-0123456789abcdef0123456789abcdef",
+    "name": "Raft notes",
+    "model": "reasoning",
+    "turn": 1
+  },
+  "message": {
+    "text": "The answer...",
+    "content": [{"type": "text", "text": "The answer..."}]
+  },
+  "artifacts": [],
+  "warnings": []
+}
+```
+
+Compatibility fields `session_id`, `name`, `model`, and `answer` are also
+returned. New integrations should use the nested fields. Raw provider data
+is excluded unless `--include-raw-response` is supplied.
+
+For long or multiline prompts:
+
+```bash
+conferllm chat --model reasoning --prompt-file ./prompt.md --json
+```
+
+Successful JSON results are written to stdout. Runtime failures exit nonzero
+and write a `conferllm.error.v1` envelope to stderr; library logging is suppressed
+in JSON mode so the envelope remains parseable. Argument-parser failures use
+the normal CLI usage error and exit code 2.
+
+## Continue and find conversations
+
+Use the actual session ID returned by the preceding call:
+
+```bash
+conferllm chat \
+  --session 20260905-0123456789abcdef0123456789abcdef \
+  --prompt "Now compare it with Paxos." \
+  --json
+```
+
+Exactly one of `--model` and `--session` is required. A session retains its
+model and name; `--name` is only accepted when creating a conversation.
+ConferLLM restores prior successful turns automatically.
+
+Find conversations using metadata:
+
+```bash
+conferllm sessions list
+conferllm sessions list --query raft --json
+conferllm sessions list --model reasoning --since 2026-09-01 --limit 20 --json
+```
+
+Filters combine with AND semantics. `--query` matches the ID or name without
+case sensitivity; model aliases match exactly. `--since` and `--until` are
+inclusive creation dates. The default limit is 50; `--limit 0` is unlimited.
+Listing never searches message bodies and reports corrupt headers as warnings
+without hiding unrelated valid sessions.
+
+For a model comparison, create one independent session per model using the
+same prompt. Attribute each answer and retain each session ID; a conversation
+does not switch models.
+
+## Work with images
+
+Repeat `--image` to preserve input order:
+
+```bash
+conferllm chat \
+  --model vision \
+  --prompt "Compare these screenshots." \
+  --image ./before.png \
+  --image ./after.png \
+  --json
+```
+
+Accepted images are copied into private, session-owned storage. Follow-ups
+use those copies even if the original files change or disappear. PNG, JPEG,
+GIF, WebP, BMP, and SVG MIME types are detected from content rather than file
+extensions.
+
+Declared capabilities are checked before reading images or calling the
+provider. Application limits bound new attachments; a model's
+`max_input_images` also counts images replayed from history. Unknown image
+capability produces a warning, not a claim of provider support.
+
+Models that return images can produce multiple ordered output artifacts,
+including image-only replies. `message.content` contains their references and
+`artifacts` contains their MIME type, size, SHA-256, local path, and stable
+`conferllm://sessions/.../artifacts/...` URI.
+
+Use `--image-output-dir PATH` for additional exported copies. If export fails,
+the canonical artifacts and successful conversation remain available. Preserve
+the session ID and handle the warning instead of repeating the model call.
+
+Generated images must be embedded data URLs. ConferLLM does not download
+remote-only image outputs, execute provider tool calls, or silently discard
+unsupported content.
+
+## Install the Agent Skill
+
+The source layout is:
+
+```text
+skills/conferllm/
+├── SKILL.md
+├── agents/openai.yaml
+└── references/
+    ├── installation.md
+    ├── multimodal.md
+    └── errors.md
+```
+
+[skills/conferllm/](skills/conferllm/) is the single source of truth for the Skill.
+The wheel embeds the same files; the source distribution preserves this layout.
+Source and installed Skills use identical relative reference paths.
+
+Install the version bundled with the executable:
+
+```bash
+conferllm skill install
+conferllm skill install --target codex
+conferllm skill install --destination /custom/skills/conferllm
+```
+
+Defaults are `~/.agents/skills/conferllm` and `~/.codex/skills/conferllm`. Installation
+is idempotent. A different existing installation is preserved unless `--force`
+is explicitly requested. Home, workspace, package directories, and their
+ancestors are never valid replacement targets.
+
+An agent invoking `$conferllm` discovers configured aliases, retains session IDs,
+preserves image order, and attributes answers. It uses the CLI rather than
+opening credential or session files.
+
+## Storage and reliability
+
+Sessions are UTF-8 JSONL files below `~/.conferllm/sessions/YYYY/MM/DD/`. Every
+committed turn contains a complete user/assistant pair and its artifact
+metadata. Directories use mode `0700`; session and image files use `0600`.
+The configuration file remains user-managed.
+
+Continuation holds a thread/process lock across loading, model execution, and
+atomic JSONL replacement. Stored image size/hash checks precede replay.
+Failed provider calls do not append a partial turn.
+
+JSONL is the commit record. A locked continuation can recover the next
+uncommitted image directory left by a crash; it never removes committed turns.
+Early staging leftovers are not automatically swept. This is not an
+exactly-once guarantee for provider calls, and long histories are not silently
+summarized or truncated.
+
+Session content can be sensitive. Keep credentials and conversations out of
+version control. Configuration/provider errors do not echo raw input or
+provider exception payloads.
+
+## Development
+
+Use `uv>=0.12.9,<0.13`:
+
+```bash
+uv sync --extra dev
+uv run ruff format --check .
+uv run ruff check .
+uv run mypy src/
+uv run pytest
+uv lock --check
+uv build
+git diff --check
+```
+
+Tests isolate the user home and block unmocked provider calls. They cover
+persistence, process/thread locking, fault recovery, protocol integration,
+Skill installation, and wheel/source-distribution content.
+
+Key dependency major versions are bounded, and `uv.lock` records the development
+resolution. Python 3.10 uses the verified LiteLLM 1.97.x line; Python 3.11+
+permits LiteLLM 1.x from 1.99.0. The project uses uv's centralized environment
+support so editable installs do not depend on hidden `.pth` files in synced
+macOS folders.
+
+## License
+
+ConferLLM is released under the [MIT License](LICENSE).