Agent Architecture Audit
Diagnoses why an AI agent app misbehaves across a 12-layer stack and proposes code-first fixes
- Repository
- affaan-m/ECC
- Stars
- 295,127
- Licence
- MIT
- Files
- 1 file
What it does
This skill teaches an AI coding agent to act as a diagnostic auditor for applications built on LLMs or autonomous agent loops. When the app gives wrong, inconsistent, or degraded answers, the agent walks through a 12-layer model of the stack (system prompt, memory, tool selection, tool execution, output formatting, platform rendering, hidden retry loops, and so on), searches the codebase with grep-style patterns for risky designs, and reports findings ranked by severity with evidence references, confidence scores, and an ordered fix plan that prioritises enforcing rules in code rather than in prompt text.
Great for
- Developers who have an agent or LLM feature that works in a playground but breaks inside their wrapper or app
- Teams investigating reports that an agent has gotten worse after a recent prompt, tool, or memory change
- Engineers preparing an agent application for production release who want a structured pre-release review
Skip it if
- General bug fixing or runtime failures such as loops, timeouts, or state errors; the skill itself points to agent-introspection-debugging for these, so it is the wrong tool there
- Security vulnerability scanning; it only mentions security-review for that purpose and does not perform a security audit itself
- Non-developers without access to the codebase, logs, and configuration, since the audit depends on reading source files and traces
Try asking
- “Audit my agent app: users say it worked last week but now ignores the tool it is supposed to call.”
- “Our chatbot's answers look correct in the logs but come out garbled in the UI. Find where the output is changed.”
- “Check my agent codebase for hidden retry or fallback LLM calls that run after the main answer.”
- “Review our memory layer for stale topics leaking into new conversations and suggest fixes.”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
agent-architecture-audit.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip agent-architecture-audit.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip agent-architecture-audit.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Give the agent concrete symptoms, the time window when they started, and the entry point users use, since the skill asks for these during scoping.
- Ask for evidence references in file:line form so each finding can be checked by hand before changes are made.
- Treat the output as a hypothesis list: the confidence scores are self-assigned, so verify the top critical findings against real traces.
What's inside
- SKILL.md
Redistributed unmodified from affaan-m/ECC under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.