Spec a change. Read it on your phone through Checkpoint. Escapement queues it overnight, Interlock ships it, and you wake up to a commit. Shippable Skills is the sibling, the reusable skills extracted from doing this work.
01
Live · v1.0.2
Interlock
Autonomous spec-driven development for Claude Code, layered on OpenSpec.
One human checkpoint. Spec a change, read it, then ship it start-to-commit with parallel agents.
A Claude Code plugin plus zero-dependency Node CLIs. /interlock:spec explores, generates proposal, design, tasks and delta-specs, reviews them, then stops for one human read. /interlock:ship then runs dependency-ordered waves of parallel agents, verifies, and commits with no mid-run input. Deterministic gate spine, codebase knowledge graph, interlock doctor preflight, PreToolUse guards that fail open outside a run, and a multi-host runner covering Claude Code CLI, ACP agents, Codex and Qwen.
JavaScript · Node ≥18 · zero deps · MIT · npm: @renzrollon/interlock
- agentic-ai
- agentic-workflow
- claude-code
- harness-engineering
- openspec
- multi-agent
- code-review
- knowledge-graph
View on GitHub ↗ (opens in a new tab)
02
Live · v0.1.0
Shippable Skills
Host-neutral agent skills for Cursor, Copilot, Codex, and Claude Code, each linted, evaluated, security-scanned, and install-verified before release.
A curated skills repo. Ships session-retro, which scores the conversation you are still in, flags token waste, and folds repeatable bits back into the skills that session called, and product-spine, engineer-as-product-owner, threading from conversations to a one-pager to epics and briefs, repo-native. Quality gate: validate with skill-lint, evaluate with skill-up at 4/4 cases, security scan with SkillSpector, then release an archive verified installable.
Install: npx skills add renzrollon/shippable-skills · MIT
- agent-skills
- cursor
- copilot
- codex
- claude-code
- productivity
- token-economy
View on GitHub ↗ (opens in a new tab)
03
In progress
Escapement
Drains a queue of approved OpenSpec changes overnight.
Starts each Interlock ship in a background Claude Code session, classifies every stop from what the run wrote to disk, and pushes one message per stop to your phone. Decides nothing about a run, only whether a change may start, when a run has stopped, and what to tell you. Node ≥22, zero runtime deps, macOS LaunchAgent supervision.
View on GitHub ↗ (opens in a new tab)
04
In progress
Checkpoint
The phone-sized inbox for the one decision an unattended Interlock run cannot make.
Reads a change from a GitHub branch, renders proposal, design, tasks and specs for a 375px screen, lets the owner Approve or Send back, and records the decision as a commit on the same branch. The git host is the message bus, so there is no database. Next.js and TypeScript.
View on GitHub ↗ (opens in a new tab)