Back to directory

espalier-engineering

A Claude Code plugin that discovers a codebase's actual conventions and encodes them as rules, skills, sub-agents and enforcement hooks, so generated code follows those conventions from the first attempt.

Visit website (opens in a new tab)junhanliu-dev.github.io
espalier-engineering

espalier-engineering reads an existing codebase, discovers the conventions already in it, and writes them down as a per-project directory of always-loaded rules, phase-loaded skills, sub-agent definitions, a wiki and programmatic hooks, so an AI coding agent's output fits the project's actual patterns instead of generic choices.

A one-time init command runs parallel scout agents to extract those unwritten conventions; after that, every change goes through a ten-stage pipeline where a coder agent writes the change, a separate reviewer agent with a different tool set checks it against the same rules, and hooks block a push that skips the gates. A drift-detection step flags documentation the code has since outgrown for review rather than letting it go silently stale.

It installs as a Claude Code plugin marketplace entry, with commands for a slimmer bug-fix lane, read-only question answering sourced from the project's own docs, a repository-wide security audit, a simplification survey and larger multi-session planning.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.