agentleFS
D-SheyD-Shey/autoharness/llms.txt

autoharness

llms.txt10 starsChanged todayMITView on GitHub

AutoHarness is an empirical harness for AI coding agents (Gemini, Claude Code, Antigravity, Cursor, Codex, Kimi Code). It automates dynamic tool discovery, security vetting, sandbox containment, and empirical verification with independent critic reviews. AutoHarness enables AI agents to start almost bare and build the exact harness each task requires: references, documentation, tools (MCP servers and skills), and empirical verification checks. All installed tools are scanned for prompt injection and data exfiltration, pinned in lockfiles, and cleanly torn down or persisted…

The file

Files
llms.txt
# AutoHarness

> AutoHarness is an empirical harness for AI coding agents (Gemini, Claude Code, Antigravity, Cursor, Codex, Kimi Code). It automates dynamic tool discovery, security vetting, sandbox containment, and empirical verification with independent critic reviews.

AutoHarness enables AI agents to start almost bare and build the exact harness each task requires: references, documentation, tools (MCP servers and skills), and empirical verification checks. All installed tools are scanned for prompt injection and data exfiltration, pinned in lockfiles, and cleanly torn down or persisted as recipes and lessons across projects.

## Core Documentation

- [Project Overview](README.md): Comprehensive guide to AutoHarness philosophy, core workflow, architecture, and security model.
- [Installation and Agent Setup](docs/installation.md): Instructions for configuring AutoHarness with Gemini, Claude Code, Antigravity, Cursor, Codex, Kimi Code, and shell hooks.
- [CLI Reference](docs/reference.md): Detailed documentation for all commands (`init`, `brief`, `need`, `search`, `vet`, `add`, `verify`, `review`, `critic`, `teardown`).
- [Security and Trust Model](docs/security.md): Threat model, static scanner rules, sandbox containment, network egress restrictions, and veto policies.
- [Contributing Guide](CONTRIBUTING.md): Code standards, zero external runtime dependency rules, development workflows, and testing.

## Design Specifications

- [Architecture and Design Principles](docs/design/README.md): Core design decisions, minimal mechanisms, and task lifecycle guarantees.
- [MCP Gateway Architecture](docs/design/specs/2026-09-21-gateway-design.md): Gateway proxying, runtime tool injection without restarts, and permission isolation.
- [Dynamic Generated Skills Design](docs/design/specs/2026-09-21-generated-skills-design.md): Architecture for generating and verifying on-the-fly skills when registry candidates do not cover requirements.

## Key Capabilities and Concepts

- [Task Modes](docs/reference.md): Task lifecycle modes (`light` for chores, `full` for capability-driven tasks, `reviewed` for mandatory independent critic reviews).
- [Recipes and Lessons](docs/reference.md): Persistent harness patterns (`recipes`) and cumulative operational insights (`lessons`) carried across projects.
- [AST Outlining](src/autoharness/vet.py): Structural compaction of large code files preserving imports, classes, and method signatures while folding function bodies to defeat hidden payload attacks.
- [Empirical Verification](docs/reference.md): Verifiable evidence collection, command execution logging (`verify run`), and independent critic reviews (`critic`).

## Optional

- [Changelog](CHANGELOG.md): Version history, new capabilities, and evolutionary changes.
- [Security Disclosure Policy](SECURITY.md): Vulnerability reporting procedures and response commitments.
- [Full Documentation Compilation](llms-full.txt): Complete, concatenated repository documentation in a single file for full-context LLM ingestion.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.