Skip to content

Repository files navigation

Model Implementation Engine v0.2.0

CI

MIE is a local implementation-continuity engine for coding agents. It preserves reviewed project goals, task boundaries, decisions, blockers, checkpoints, and evidence metadata in one SQLite ledger without editing source, executing project commands, scheduling workers, or calling model providers.

Quick start

Requires Python 3.11 or later.

git clone https://github.com/Zerolitter/MIE-Model-Implementation-Engine.git
cd MIE-Model-Implementation-Engine
python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -e .
.\.venv\Scripts\python.exe scripts\verify.py

On macOS or Linux, replace .\.venv\Scripts\python.exe with ./.venv/bin/python.

Start a project ledger

  1. Copy examples/mie-build.seed.json to <PROJECT_ROOT>/mie.seed.json.
  2. Replace its example project, decisions, tasks, and history with reviewed project data.
  3. Initialize a separate local ledger with the virtual-environment interpreter.

Windows PowerShell:

$MiePython = (Resolve-Path .\.venv\Scripts\python.exe).Path
$ProjectRoot = 'C:\path\to\project'
$Seed = Join-Path $ProjectRoot 'mie.seed.json'
$Database = Join-Path $ProjectRoot '.mie\state.sqlite3'
Copy-Item examples\mie-build.seed.json $Seed
& $MiePython -m mie --db $Database init `
  --root $ProjectRoot `
  --seed $Seed `
  --reviewed
& $MiePython -m mie --db $Database resume

macOS or Linux:

MIE_PYTHON=\"$PWD/.venv/bin/python\"
PROJECT_ROOT=/path/to/project
SEED=\"$PROJECT_ROOT/mie.seed.json\"
DATABASE=\"$PROJECT_ROOT/.mie/state.sqlite3\"
cp examples/mie-build.seed.json \"$SEED\"
\"$MIE_PYTHON\" -m mie --db \"$DATABASE\" init \
  --root \"$PROJECT_ROOT\" \
  --seed \"$SEED\" \
  --reviewed
\"$MIE_PYTHON\" -m mie --db \"$DATABASE\" resume

Add .mie/ to the target project's ignore rules. Keep each project's ledger, backups, and evidence files private.

Optional MCP integration

Install the portable MCP extra with the same interpreter:

.\.venv\Scripts\python.exe -m pip install -e \".[mcp]\"
./.venv/bin/python -m pip install -e '.[mcp]'

requirements-mcp.lock records the exact dependency set used for release verification when reproducibility matters more than accepting compatible updates.

Copy examples/omp.mcp.json into the target project's .omp/mcp.json, replace its placeholders with absolute paths, then reload the MCP configuration. The adapter exposes nine bounded stdio tools:

  • Read: mie_resume, mie_status, mie_next, mie_ready, mie_brief, mie_review
  • Record workflow state: mie_checkpoint, mie_submit, mie_propose

MCP cannot initialize or migrate a ledger, apply a proposal, accept a submission, create a backup, run SQL or shell commands, edit source, schedule work, or call a provider.

v0.2 capabilities

  • One-call recovery of the current proposal, submission, blocker, implementation task, contract reconciliation, or define-work state.
  • Dependency-ready queues and exact one-task readiness checks with machine-readable exclusions.
  • Revision-bound checkpoints, structured blockers, submissions, local review, and acceptance history.
  • Atomic 3–10 task work slices and project-contract supersession with explicit stale-task reconciliation.
  • Required-reference hash binding, bounded canonical payloads, idempotent writes, and stale-revision rejection.
  • Reviewed schema-v1 to schema-v2 migration with a required integrity-checked rollback backup.
  • Immutable engine and tool-contract identity in CLI and MCP status/resume responses.
  • Content-free diagnostics that keep core, stdio, host, and model timing layers separate.

Trust boundary

--reviewed and actor labels are workflow acknowledgements, not authentication. Evidence hashes establish file identity, not correctness or runner identity. SQLite is local and unencrypted. The consuming agent host may forward MIE responses to its configured model provider.

MIE is not a source index, worker lease, scheduler, model router, dashboard, network service, sandbox, or independent reviewer.

Documentation

  • RELEASE_NOTES.md — v0.2 changes, upgrade path, and verified boundaries
  • docs/01_BUILD_CONTRACT.md — data model, invariants, bounds, and authority
  • docs/02_FINISH_AND_ACCEPT.md — acceptance scenarios
  • docs/03_OPERATIONS_AND_MCP.md — CLI, migration, backup, and MCP operation
  • docs/04_EXPANSION.md — explicitly deferred capabilities
  • STATUS.md — concise capability matrix

License

Licensed under either the Apache License 2.0 or the MIT License, at your option. See LICENSE-APACHE and LICENSE-MIT.

About

MIE is a local implementation-continuity engine with a dependency-free Python core, a CLI, and an optional stdio MCP adapter. It retains reviewed goals, task boundaries, decisions, checkpoints, and evidence metadata without executing project work or calling model providers.

Topics

Resources

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages