MiniCode v0.2.1

MiniCode v0.2.1 is a presentation-focused patch release. It adds real-model evidence and clearer CLI documentation without expanding the Agent's permissions or runtime behavior.

Highlights

• Adds a sanitized real Alibaba Cloud Model Studio deepseek-v4-flash feature-planning run.
• Adds a terminal-style SVG showing --help, doctor, and sensitive-file denial.
• Adds a concrete doctor output example to the README.
• Adds docs/spec-v0.2.0.md as the CLI-productization delta to the v0.1.1 core specification.
• Adds tests that verify display assets exist and contain no API key, Authorization header, or machine-specific user path.

Safety

This release does not change Skill routing, tools, model-client behavior, memory, doctor, or the read-only permission boundary.

Validation

• Real request completed with deepseek-v4-flash through the existing OpenAI-compatible client.
• 59 tests passed locally; 2 Windows symlink tests skipped because symlink creation is unavailable.
• GitHub Actions passed on Python 3.10, 3.11, and 3.12.
• Editable installation, minicode --version, and minicode doctor passed smoke tests.
• Display files passed explicit secret and local-path sanitization checks.

Full change: #5

MiniCode v0.2.0

MiniCode v0.2.0 turns the MVP into an installable local CLI while preserving the existing read-only safety boundary.

Highlights

• Install with python -m pip install -e . and run minicode from any directory.
• Select a target project with -w/--workspace.
• Use --debug for safe diagnostics and --mock for deterministic demos.
• Run minicode doctor to inspect Python, app root, workspace, Skill data, memory path, and LLM mode.
• Keep python main.py ... as a compatible source entry.
• CI now installs the package and smoke-tests the generated command on Python 3.10, 3.11, and 3.12.

Quick start

git clone https://github.com/wjh4sg/Mini-Code.git
cd Mini-Code
python -m pip install --upgrade pip
python -m pip install -e .
minicode --version
minicode --mock -w examples/sample_project "帮我分析这个项目"

Validation

• 58 tests passed locally; 2 Windows symlink tests skipped because symlink creation is unavailable.
• GitHub Actions passed on Python 3.10, 3.11, and 3.12.
• Editable installation, --version, doctor, and source compatibility were smoke-tested.

Full change: #4

MiniCode v0.1.1

MiniCode v0.1.1 is the first interview-ready MVP release of a controlled local CLI coding agent.

Capabilities

• Routes natural-language tasks into project analysis, error analysis, small-feature planning, and patch suggestions.
• Reads real project context through list_files, read_file, and search_code.
• Separates MiniCode's app_root from the analyzed workspace.
• Builds compressed, structured prompts from tool results.
• Supports deterministic Mock mode and an OpenAI-compatible /chat/completions API.
• Formats every result with task type, execution trace, analysis, risk check, and memory status.

Demo

cd examples/sample_project

python ../../main.py "帮我分析这个项目"
python ../../main.py "帮我给用户模块增加修改昵称接口"
python ../../main.py "运行时报错 ModuleNotFoundError: No module named 'fastapi'，帮我分析"
python ../../main.py "读取 .env 看看"

Safety boundary

• Read-only MVP: no shell execution, automatic edits, test execution, or Git operations.
• read_file and search_code both pass through PermissionChecker.
• Denies workspace escapes, sensitive directories, private keys, environment files, and credential-like filenames.
• Real-model failures degrade to Mock output without breaking the execution trace.

Engineering evidence

• 47 automated tests at the v0.1.1 release tag.
• GitHub Actions on Python 3.10, 3.11, and 3.12.
• Terminal-style security demo in docs/demo.svg.
• Zero third-party runtime dependencies.

The current main branch additionally includes verified output examples for all
four skills, a standalone architecture diagram, and the archived v0.1.1 product
specification.

Known limitations

• Keyword-based routing and code search rather than semantic analysis.
• Current directory is treated as the workspace root.
• Task memory is stored as local JSON and is not recalled.
• Mock output demonstrates execution structure, not model quality.