templetongroup

AI Structure Audit — Security skill for Claude Code

Security community

Evidence-driven audit of AI systems across prompt, context, harness, loop, and graph layers.

How to install AI Structure Audit

This entry records only its repository, not the path inside it, so there is no exact command to give. Open templetongroup/ai-structure-audit and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Structure Audit does

Evidence-driven audit of AI systems across prompt, context, harness, loop, and graph layers.

Alternatives in Security

Octoboss Reorganize Tentacles — You are the Octoboss — a cross-tentacle orchestrator 1.4k ★
Imcodes — The IM for agents 960 ★
Maestro Knowledge — Intent-driven knowledge-store and Run knowledge lifecycle management — audit/prune, stage candidates (with sig 530 ★

README

AI Structure Audit

AI Structure Audit is an evidence-driven skill for reviewing AI assistants, agents, RAG systems, tool-using workflows, and multi-agent architectures.

It audits five nested layers:

Prompt
Context
Harness
Loop
Graph

The skill diagnoses from the bottom up, identifies the lowest failing layer, recommends the smallest effective fixes, and offers to implement those fixes only after approval.

What it checks

prompt clarity, constraints, examples, output contracts, and evaluations
context visibility, retrieval, memory, provenance, freshness, and token discipline
tool schemas, permissions, sandboxing, retries, idempotency, verification, and observability
verifier quality, stop conditions, no-progress detection, budgets, and convergence
node responsibilities, routing, state flow, parallelism, aggregation, and graph justification
common cross-layer misdiagnoses, such as adding agents to compensate for a weak verifier

Install

Clone the repository into your agent's skills directory:

git clone https://github.com/templetongroup/ai-structure-audit.git ~/.codex/skills/ai-structure-audit

For OpenClaw, install it through your normal reviewed skill-install process and verify visibility with `openclaw skills check`.

Use

Examples:

Use $ai-structure-audit to review this agent system.
Audit this RAG workflow and tell me which structural layer is actually failing.
Review this multi-agent architecture before we add another node.
Run an AI Structure Audit, suggest fixes, and offer to implement them.

Audit mode is read-only. Implementation requires explicit approval after the report identifies the proposed scope, affected systems, validation plan, risk, and rollback.