skills.rest
GuiomeB
GuiomeBCOMMUNITY
Explore all skills by this author →
GitHub

simple-ai-harness-blueprint

Generates and audits AGENTS.md-based AI collaboration harnesses for code repositories.

Updated May 16, 2026
Search Tags:#claude code#scaffold#agents.md#harness#code repository#ci validation#ai coding agents
One-click install
npx skills add https://github.com/GuiomeB/simple-ai-harness-blueprint --skill simple-ai-harness-blueprint-guiomeb
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: simple-ai-harness-blueprint
Source: https://github.com/GuiomeB/simple-ai-harness-blueprint
Command: npx skills add https://github.com/GuiomeB/simple-ai-harness-blueprint --skill simple-ai-harness-blueprint-guiomeb

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? AI coding agents forget rules mid-session, config files sprawl and contradict each other, and stale doctrine drifts from the actual codebase. This Skill installs a structured collaboration harness (AGENTS.md, CLAUDE.md, .agents/ directory) that keeps agents and humans aligned without overwriting existing project memory. ## Core Features & Use Cases - Bootstrap or audit a harness: Inventory an existing repo, infer the right size (S/M/L) from signals like agent count and AGENTS.md line count, and generate only the missing files from templates. - Three additive sizes plus L+ profile: S ships the minimal contract and learning loop; M adds a router, status log, and PR risk-rail template; L adds patterns, rules, ADRs, a meta-validator, and CI rail-guard; L+ is an opt-in ADR-gated autonomy profile with hooks and a bounded /loop workflow. - Drift detection and learning loop: Ships a meta-validator script, a fleet audit script, and /learn and /retro workflows that turn each friction into one actionable edit. - Use Case: Point it at a repo where Claude Code keeps ignoring conventions; it detects the existing .cursorrules and .github folder, proposes size M, and generates AGENTS.md, ROUTER.md, and a PR template without touching existing files. ## Quick Start Audit my repository against the AI harness blueprint and tell me which files are missing for size M.

Frequently Asked Questions about simple-ai-harness-blueprint

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I set up AGENTS.md for AI coding agents in my repo?▼

Run the skill against your repository and it inventories existing files, infers a size (S, M, or L) from signals like agent count and AGENTS.md length, then generates only the missing files from templates. It never overwrites existing doctrine silently.

What is the difference between size S, M, and L harnesses?▼

S is a minimal 4-5 file setup with the core contract and learning loop. M adds a router, status log, and formalized PR risk-rail template. L adds patterns, rules, ADRs, a meta-validator script, and machine-enforced CI rail gating via CODEOWNERS.

Does this work with Claude Code, Cursor, and Codex CLI?▼

Yes. AGENTS.md is the canonical contract read natively by Codex CLI and Cursor, while thin adapter files like CLAUDE.md restate the load order for tools that need them. The same SKILL.md works for both Claude and Codex Desktop.

Will it overwrite my existing CLAUDE.md or AGENTS.md?▼

No. On repos with existing agent doctrine it proposes an additive merge, inserting universal sections like the Karpathy rules and load order only where missing. It adapts to alternate layouts such as .cursor/rules/ instead of forcing a parallel structure.

When should I not use this AI harness blueprint?▼

Skip it for product code changes, PR reviews, IDE configuration, or generic project scaffolding. It operates only on the documentation surface agents read and does not touch application code, run tests, or decide architecture.

What is the L+ autonomy profile and when do I need it?▼

L+ is an opt-in profile on top of L that adds a /loop workflow with hard brakes, .claude/ hooks, and a headless runner for unattended execution. Adopt it only when all five signals hold, including real unattended runs and genuine budget needs, and record the decision in an ADR.

Related Skills

Explore More Software Engineering Skills→
johnpapajohnpapa
120

ai-ready

Analyze a repository to generate AI-ready configuration assets and a readiness report.

Community
Intermediate
sandeepmvlsandeepmvl
21

ci-cd-github-actions-rails

Generate GitHub Actions CI/CD workflows for Rails 8 with RSpec, RuboCop, Brakeman, and Kamal.

Community
Advanced
KristopherlbKristopherlb

retrospective

Create structured checkpoints and retrospective analyses to document project learnings and improvements.

Community
Advanced
flightctlflightctl
6

ai-ready

Scan codebase structure and convention files to generate or refresh AGENTS.md.

Official
Intermediate