mira-process 0.1.1

Modular CLI and LLM tool engine for project operations

MIRA-ENGINE

MIRA-ENGINE is a modular Python CLI for project-aware file operations and LLM-friendly tool wrappers.

Commands

AI-facing commands:

• mira help or any unknown command prints only the visible AI-facing command list instead of trying to index a path.
• mira create creates files or folders. Interactive file creation asks for the file description only after CONTINUE or after the immediate edit flow.
• mira delete removes files or folders after confirmation, using root-style paths for root-style input and absolute paths only for absolute input.
• mira edit replaces selected lines in a file, with optional explore checks and targeted comment or description updates for the affected ranges.
• mira project prints the project tree with root files listed before folders. Empty workspaces show Use <tool_use>create</tool_use> guidance.
• mira explore analyzes a file, offers VIEW / EDIT / EXIT / ADD OR EDIT DESCRIPTION, and only asks for chunk comments when the full chunk range was actually viewed.
• mira memory and mira plan manage .mira/memory.md and .mira/plan.md.
• mira terminal runs commands through one of two persistent PowerShell-backed terminals.

Admin commands:

• mira . or mira sync [path] updates the stateful index and only reprocesses changed files.
• mira configs [view|init|set|reset] views or updates mira.config.json.
• mira index add <path> and mira index remove <path> manage manual include or exclude rules.
• mira history shows grouped session history versions like V1, V2, and so on.
• mira diff [version] [other_version] compares file versions. mira diff --history <version> shows a compact + / - batch summary for one history version.
• mira restore <version> restores one file version. mira restore --history <version> restores an entire recorded history version as a batch.

LLM wrappers

Use --llm to wrap output in configurable tags from mira.config.json:

<tool_use>
create
path: notes.txt
</tool_use>
<tool_response>
Created file: notes.txt
</tool_response>

Chunking

Python files use AST-aware structure extraction. Other files use fallback chunking from languages/default/analyzer.py, targeting 80 lines with chunk boundaries selected at nearby blank lines up to 200 lines, then falling back to 100 lines when no natural boundary is found.

Stateful Index

Running mira . materializes a lightweight project state under .mira/:

• .mira/project.txt stores the tracked project tree with file descriptions.
• .mira/commands.txt stores the command catalog with one-line descriptions.
• .mira/files/<path>/skeleton.txt stores the explore-style file summary.
• .mira/versions/<path>/<version>.txt stores the first retained snapshot for a file.
• .mira/versions/<path>/<version>.ndiff stores later retained revisions as independent ndiff payloads instead of full snapshots.

History is also grouped into retained history versions so one sync or restore batch can be inspected and replayed as a unit.

Descriptions and structure comments are manual metadata. They are added or updated through interactive create, explore, and edit flows instead of being autogenerated from AST docstrings or fallback heuristics. Skeleton output omits empty description, dependency, and comment metadata to keep low-noise files compact.

The sync path skips unchanged files by mtime, size, hash, and artifact presence, so rerunning mira . stays cheap when only a few files changed.