Analyze Agent Codebase

Analyze Agent Codebase Skill

When to use this skill

Use this skill when the user wants to perform a comprehensive architectural analysis of an agentic AI codebase.

Instructions

Analyze the target codebase in four phases.

Each cluster has its own questions file in the skill directory (e.g. questions_cluster_a.md).

Global Token Rules

• Do NOT launch parallel subagents.
• Run clusters sequentially.
• Do NOT read the entire repository per cluster.
• Read ONLY the assigned cluster questions file.
• Read ONLY files relevant to the cluster questions.
• Index-driven lookup only — when a section says "Find" or "Find and read", do not search the filesystem. Use the Key File Registry in index.md (already in context from Cluster A) to identify candidate files, then read only those. If a topic has no matching registry entry, write: Not indexed — verify Phase 1 captured all relevant files.

Global Output Rules

• Iventories: always use tables (tools, models, entities, tests)
• Findings with evidence: always use bullet points in format - [finding] — evidence: path/to/file.ext:Lxx
• Comparisons or yes/no pattern checks: always use a table with columns: Pattern | Implemented | Evidence
• Evidence format: path/to/file.ext:Lxx-Lyy
• NO prose or narrative explanations (except Section 14: Decision Extraction and Classification).
• NO restating questions.
• NO long evidence excerpts.
• Do NOT duplicate findings across sections.
• Write each cluster file immediately after completion.

Per-section Output Format:

• Provide Finding: what is implemented in the codebase.
• Provide Evidence: exact file path and line number (file.py:Lxx) with a direct code quote.
• If nothing is implemented, write exactly: No implementation.
• If partially implemented, write: Partial — [what exists] / [what's missing].
• Separate implemented-and-active, implemented-but-disabled, and not-implemented items.
• Do not speculate. Only use code-level evidence.

Output all results to {CODEBASE_ROOT}/_analysis/ directory inside the target codebase root.

Phase 1: Index

Build a structural map of the codebase. This must complete before Phase 2.

1. List the full directory tree (exclude nodemodules, .git, _pycache__, .venv, venv, dist, build, .next).
2. Read anchor files if they exist:

• Dependency manifests: pyproject.toml, requirements.txt, setup.py, package.json, go.mod, Cargo.toml
• Entry points: main.py, app.py, index.ts, server.py, or files referenced in manifest scripts/entry_points
• Configuration: config.py, settings.py, docker-compose.yml, Dockerfile (must not read .env)
• CI/CD: .github/workflows/*, Jenkinsfile, .github/workflows

1. Extract: orchestration framework and version, LLM provider SDKs, and key infrastructure dependencies.
2. Identify locations of: agent/graph definitions, tool definitions, state/memory definitions, tests, and deployment config live.
3. Write {CODEBASE_ROOT}/_analysis/index.md. No analysis. Mapping only.

Structure: ### Project Metadata

• Project name
• Stated purpose: Quote the package description or README opening verbatim
• Use case domain: e.g. customer service, coding assistant, research, data analysis, DevOps
• Project type: Product / internal tool / demo / research prototype
• Author(s) or organization

### Framework

• Name:
• Version:
• Provider SDKs:

### Directory Map Top-level directory table:

| Directory | Purpose |

### Key File Registry

| Category | File Path | Purpose |

Phase 2: Clustered Analysis

Run clusters one at a time. Write file before proceeding to next.

Cluster assignments:

| Cluster | Questions file | Output file (written by main agent) | |---|---|---| | A: Core Architecture | questionsclustera.md | clusteracorearchitecture.md | | B: Execution, State, Memory | questionsclusterb.md | clusterbexecutionstatememory.md | | C: Tools and Retrieval | questionsclusterc.md | clusterctoolsretrieval.md | | D: Data and Adaptation | questionsclusterd.md | clusterddataadaptation.md | | E: Safety and Security | questionsclustere.md | clusteresafetysecurity.md | | F: Ops | questionsclusterf.md | clusterfops.md |

Cluster Prompt Template:

Use this template for each cluster:

> Read {CLUSTER_QUESTIONS_FILE}. > Identify relevant files using the Key File Registry in index.md (already in context). > Read only those files. > Produce structured findings only.

After each cluster completes, write output to: {CODEBASE_ROOT}/_analysis/{OUTPUT_FILE}. Then release: stop referencing this cluster's question file and the source files read for it — do not re-read them. Keep _analysis/index.md in context. Then proceed to the next cluster.

Phase 3: Synthesis

After all cluster files are written:

1. All cluster output files and index.md are already in context — do not re-read them from disk.
2. Read questions_synthesis.md.
3. Do not repeat findings from cluster files — reference them by section number.
4. Write output to {CODEBASE_ROOT}/_analysis/synthesis.md

Phase 4: Assemble

Run this shell command to produce {CODEBASE_ROOT}/_analysis/full_report.md — do not read or rewrite the cluster files yourself:

{
  printf "# Agent Codebase Analysis Report\n**Codebase**: {CODEBASE_ROOT}\n**Date**: {DATE}\n\n"
  cat \
    {CODEBASE_ROOT}/_analysis/cluster_a_core_architecture.md \
    {CODEBASE_ROOT}/_analysis/cluster_b_execution_state_memory.md \
    {CODEBASE_ROOT}/_analysis/cluster_c_tools_retrieval.md \
    {CODEBASE_ROOT}/_analysis/cluster_d_data_adaptation.md \
    {CODEBASE_ROOT}/_analysis/cluster_e_safety_security.md \
    {CODEBASE_ROOT}/_analysis/cluster_f_ops.md \
    {CODEBASE_ROOT}/_analysis/synthesis.md
} > {CODEBASE_ROOT}/_analysis/full_report.md

Error handling

• If a cluster fails or times out, re-run only that cluster.
• If the index file is missing when a cluster starts, fail immediately.
• When re-running a single cluster in isolation after failure, read index.md before starting.
• If output limit reached:
• Stop at last fully completed section.
• Append a note listing incomplete sections from the cluster questions file.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ProsusAI
• Source: ProsusAI/prism
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.