writ-vcs 0.1.1

AI-native version control for agentic workflows. Structured checkpoints, spec-driven development, and multi-agent coordination.

writ-vcs

AI-native version control for agentic development.

Structured checkpoints, spec-driven development, and multi-agent coordination for agentic AI workflows.

Install

pip install writ-vcs

Quick start

# Human sets up the project (once)
writ init

# Agents work. Each agent uses three commands:
writ spec add "task description"    # create a task
writ seal -s "what I did"           # checkpoint work
writ spec done                      # mark task complete

# Human commits everything to git (once)
writ finish
git push

That's the full workflow. writ init detects your environment and sets up everything automatically. Agents discover writ via MCP tools, slash commands, or CLI. writ finish converges overlapping work and commits to git.

Python SDK

import writ

repo = writ.Repository.open(".")
ctx = repo.context()                                    # full project state
repo.seal(summary="work done", agent_id="worker-3")    # checkpoint (auto-scoped)
repo.finish()                                           # converge + commit

CLI

writ init                                          # guided project setup
writ context                                       # full project state (one call)
writ context --format toon                         # token-optimized output
writ spec add "JWT-based auth with token refresh"  # create a spec
writ seal -s "added auth" --spec auth              # structured checkpoint
writ spec done                                     # mark spec complete
writ status                                        # fleet overview
writ finish                                        # converge + commit to git
writ log                                           # seal history
writ verify                                        # cryptographic chain validation

Features

• Structured context in one call. writ context returns specs, seals, working state, file contention, integration risk. One call, one structured response.
• Convergence engine. Merges overlapping agent work using genesis trees as base. Auto resolves non-conflicting changes. Escalates real conflicts with confidence scores.
• Cryptographic integrity. BLAKE3 hash chains and Ed25519 signatures on every seal. writ verify validates the full history.
• Spec-driven workflow. Specs are structured requirements that agents work against. Task tracking with lifecycle, agent claiming, and genesis snapshots.
• Git round trip. writ finish converges outstanding work and commits to git with full provenance.
• MCP server. Native MCP server built in Rust. 21 tools available through the protocol agents already speak.
• TOON format. Token-optimized output that saves 12% over JSON per call. Spec-scoped context saves 65% at fleet scale.

Links

• Documentation
• Repository
• Issues