Code Pathfinder— agent skill

<div align="center"> <img src="./assets/banner.png" alt="Code Pathfinder - Open-source SAST with cross-file dataflow analysis" width="100%"> </div>

<div align="center">

<h3>Open-source SAST engine that traces vulnerabilities across files and functions</h3>

Website · Docs · Rule Registry · MCP Server · Blog

</div>

Quick Start

Install:

brew install shivasurya/tap/pathfinder

Scan a Python project (rules download automatically):

pathfinder scan --ruleset python/all --project .

Scan Dockerfiles:

pathfinder scan --ruleset docker/all --project .

No config files, no API keys, no cloud accounts. Results in your terminal in seconds.

<!-- TODO: Add demo video/GIF here -->

What is Code Pathfinder?

Code Pathfinder is an open-source static analysis engine that builds a graph of your codebase and traces how data flows through it. It parses source code into Abstract Syntax Trees, constructs call graphs across files, and runs taint analysis to find source-to-sink vulnerabilities that span multiple files and function boundaries.

v2.0 introduces cross-file dataflow analysis: trace user input from an HTTP handler in one file through helper functions and into a SQL query in another file. This is the kind of analysis that pattern-matching tools miss entirely.

Cross-File Taint Analysis

Most open-source SAST tools operate on single files. Code Pathfinder v2.0 tracks tainted data across file boundaries:

app.py:5    user_input = request.get("query")     ← Source: user-controlled input
  ↓ calls
db.py:12    cursor.execute(query)                  ← Sink: SQL execution

The engine builds a Variable Dependency Graph (VDG) per function, then connects them through inter-procedural taint transfer summaries. When user_input flows into a function parameter in another file, the taint propagates through the call graph to the sink.

How It Works

Source Code → Tree-sitter AST → Call Graph → Variable Dependency Graph → Taint Analysis → Findings
                                     ↓
                              Inter-procedural
                              Taint Summaries
                              (cross-file flows)

1. Parse: Tree-sitter builds ASTs for Python, Dockerfiles, and Docker Compose files
2. Index: Extract functions, call sites, parameters, and assignments into a queryable call graph
3. Analyze: Build VDGs per function, resolve inter-procedural flows, run taint analysis
4. Detect: Python-based security rules query the graph to find source-to-sink paths
5. Report: Output findings as text, JSON, SARIF (GitHub Code Scanning), or CSV

190 Security Rules, Ready to Use

Rules download from CDN automatically. No need to clone the repo or manage rule files.

# Scan with a specific bundle
pathfinder scan --ruleset python/django --project .

# Scan with multiple bundles
pathfinder scan --ruleset python/flask --ruleset python/jwt --project .

# Scan a single rule
pathfinder scan --ruleset python/PYTHON-DJANGO-SEC-001 --project .

# Scan all rules for a language
pathfinder scan --ruleset python/all --project .

Browse all rules with examples and test cases at the Rule Registry.

MCP Server for AI Coding Assistants

Code Pathfinder runs as an MCP server, giving Claude Code, Cursor, Cline, and other AI assistants access to call graphs, data flows, and security analysis. More context than LSP, focused on security and code structure.

pathfinder serve --project .

The MCP server exposes tools for querying the code graph: find callers/callees, trace data flows, search for patterns, and run security rules — all available to the AI assistant during code review or development.

Write Custom Rules

Security rules are Python scripts using the PathFinder SDK. Define sources, sinks, and sanitizers — the dataflow engine handles the analysis.

Here's a real rule from the repo (PYTHON-DJANGO-SEC-001) that detects SQL injection in Django:

from codepathfinder import calls, flows, QueryType
from codepathfinder.presets import PropagationPresets

class DBCursor(QueryType):
    fqns = ["sqlite3.Cursor", "psycopg2.extensions.cursor"]
    match_subclasses = True

@python_rule(
    id="PYTHON-DJANGO-SEC-001",
    name="Django SQL Injection via cursor.execute()",
    severity="CRITICAL",
    cwe="CWE-89",
)
def detect_django_cursor_sqli():
    return flows(
        from_sources=[
            calls("request.GET.get"),
            calls("request.POST.get"),
        ],
        to_sinks=[
            DBCursor.method("execute").tracks(0),
            calls("cursor.execute"),
        ],
        sanitized_by=[calls("escape"), calls("escape_string")],
        propagates_through=PropagationPresets.standard(),
        scope="global",  # cross-file taint analysis
    )

# Run your custom rules
pathfinder scan --rules ./my_rules/ --project .

Explore all 190 rules in the rules/ directory or browse the Rule Registry. See the rule writing guide and dataflow documentation to write your own.

See the rule writing guide and dataflow documentation for more.

Installation

Homebrew (Recommended)

brew install shivasurya/tap/pathfinder

pip

Installs the CLI binary and Python SDK for writing rules.

pip install codepathfinder

Docker

docker pull shivasurya/code-pathfinder:stable-latest

docker run --rm -v "$(pwd):/src" \
  shivasurya/code-pathfinder:stable-latest \
  scan --ruleset python/all --project /src

Pre-Built Binaries

Download from GitHub Releases for Linux (amd64, arm64), macOS (Intel, Apple Silicon), and Windows (x64).

From Source

git clone https://github.com/shivasurya/code-pathfinder
cd code-pathfinder/sast-engine
gradle buildGo
./build/go/pathfinder --help

Usage

# Scan with text output (default)
pathfinder scan --ruleset python/all --project .

# JSON output
pathfinder scan --ruleset python/all --project . --output json --output-file results.json

# SARIF output (GitHub Code Scanning)
pathfinder scan --ruleset python/all --project . --output sarif --output-file results.sarif

# CSV output
pathfinder scan --ruleset python/all --project . --output csv --output-file results.csv

# Fail CI on critical/high findings
pathfinder scan --ruleset python/all --project . --fail-on=critical,high

# MCP server mode
pathfinder serve --project .

# Verbose output with statistics
pathfinder scan --ruleset python/all --project . --verbose

GitHub Action

name: Code Pathfinder Security SAST Scan

on:
  pull_request:

permissions:
  security-events: write
  contents: read
  pull-requests: write

jobs:
  security-scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
        with:
          fetch-depth: 0

      - name: Run Security Scan
        uses: shivasurya/[email protected]
        with:
          ruleset: python/all, docker/all, docker-compose/all
          verbose: true
          pr-comment: ${{ github.event_name == 'pull_request' }}
          pr-inline: ${{ github.event_name == 'pull_request' }}
          github-token: ${{ secrets.GITHUB_TOKEN }}

      - name: Upload SARIF
        uses: github/codeql-action/upload-sarif@v4
        if: always()
        with:
          sarif_file: pathfinder-results.sarif

See the full example: .github/workflows/code-pathfinder-scan.yml

<details> <summary><strong>Action Inputs</strong></summary>

Either rules or ruleset is required.

</details>

Supported Languages

Contributing

Contributions are welcome. Read the Contributing Guide for setup instructions, how to run tests locally, and the PR process.

Pushing an in-product announcement

In-product announcements (workshops, blog posts, security advisories) are managed via release/latest.json. Add an entry to announcements[], open a PR, and once it merges to main the publish workflow uploads the manifest to the CDN within ~60 seconds. See the version-update-check tech spec for the schema and version_range semantics.

All contributors must sign the Contributor License Agreement (CLA) before any pull request can be merged.

• Report bugs or request features
• Ask questions or start a discussion
• Write security rules

License

Apache-2.0