Skip to content
← Tools / Skills
Skill · .md packageData-System-School/10 trustMetrics-checked

Investigate Codebase

Evidence-backed codebase investigation instead of confident tours

Visit official site →

About

A single portable skill for orienting in unfamiliar repositories: it builds the smallest model that answers the actual question, with every claim tied to evidence or explicitly marked unknown. Four modes cover orienting, tracing runtime flows, analyzing PR blast radius, and verifying generated code, with a strict evidence ledger and real end-to-end examples against DuckDB and vLLM included.

What it can do

  • Map a large codebase into components with a focused reading route
  • Trace runtime flows as falsifiable causal paths
  • Assess PR impact with explicit source, runtime, inferred, or unknown labels

Setup

git clone https://github.com/Data-System-School/agent-skills.git

cp -r agent-skills/investigate-codebase ~/.claude/skills/

Or symlink it so git pull updates the installed skill

Prerequisites

Claude Code or OpenAI Codex

Use cases

orienting in an unfamiliar repositorytracing a runtime data flowanalyzing pr impact and blast radiusverifying generated code against oracles

Works with

Claude CodeCodex

Trust score — 6/10

/10 trustMetrics-checked

Scored 6/10 (metrics tier, metrics-v1). Evidence: Evidence-backed codebase investigation skill (10 stars, MIT) from Data-System-School; modes and install verified from the README on 2026-10-13.; Eight real end-to-end example runs against DuckDB and vLLM, including failed checks — an honest evaluation standard.; Single-skill repo; deliberately read-only unless asked otherwise, which is a strength for audits but limits hands-on fixes.. License: MIT per GitHub license field..

★ 10 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
4.0/10
Maintenance · recency, releases, issue load
5.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · Evidence-backed codebase investigation skill (10 stars, MIT) from Data-System-School; modes and install verified from the README on 2026-10-13.
  • · Eight real end-to-end example runs against DuckDB and vLLM, including failed checks — an honest evaluation standard.
  • · Single-skill repo; deliberately read-only unless asked otherwise, which is a strength for audits but limits hands-on fixes.

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.