Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
oauth-sink-check — Trace untrusted OAuth endpoint URLs from source to shell sink — the CVE-2026-119025 attack class, checked in your code, lockfile and MCP configs. | Kitploit
Tools/GitHubGitHub/shouvikkkk/oauth-sink-check
Static AnalysisVulnerability AnalysisCode AnalysisShellcodeWeb SecurityDevSecOpsSupply Chain Security
GitHubshouvikkkk/oauth-sink-check

oauth-sink-check

Trace untrusted OAuth endpoint URLs from source to shell sink — the CVE-2026-119025 attack class, checked in your code, lockfile and MCP configs.

View Repository
21 days agoNot yet reviewed
Share

oauth-sink-check

Trace your code from an untrusted OAuth/endpoint URL source to a shell or browser-open sink, simulate whether a PowerShell payload survives the quoting used, and check your lockfile and MCP configs — for the attack class behind CVE-2026-119025.

Stdlib only. No dependencies. This tool never executes payloads; it only inspects strings.

The problem

On 19 October 2026, Pillar Security disclosed that OpenCode 1.0.137–1.18.32 on Windows/WSL let an attacker-controlled remote MCP server execute arbitrary code during MCP OAuth sign-in. The chain:

server's discovery document → MCP SDK → oauth-provider.ts → mcp/index.ts → browser.ts → open() → PowerShell.

The server-supplied authorization_endpoint URL went unescaped into PowerShell. PowerShell expands $(…) inside double-quoted strings, so a well-formed https URL carrying a subexpression ran as code. Pillar's example payload:

https://evil.test/$([IO.File]::WriteAllText('C:\Users\Public\pwned.txt','RCE'))

The fix landed in OpenCode 1.18.33 (the open package's 11.0.4 plus http/https scheme checks); the vendor said malicious MCP servers were "outside their threat model". Pillar proposed CVSS 9.8.

On 29 October 2026, CVE-2026-119025 was published: the open npm package 7.2.0 before 11.0.0 allows OS command injection via crafted URLs on Windows/WSL (CWE-78, CVSS 7.5). Affected consumers include MCP Inspector before 2.12.0 (GHSA-q978-78c7-phqw) and OpenCode before 1.18.33. open 11.0.0 fixed the escaping (2025-11-15) with no security wording, so ^10 pins kept the vulnerable version for most of a year; mcp-remote 0.15.2 still declares ^10.1.0.

The gap this fills

  • npm audit / CVE feeds flag vulnerable versions only — they say nothing about whether your code actually passes an untrusted URL to a sink.
  • Generic SAST misses PowerShell quoting semantics: the payload is not a weird string, it is a valid https URL.
  • Scheme allowlists (https-only) do NOT stop it — the payload rides inside a valid https URL. An https check is necessary but not sufficient; the URL must also be encoded/escaped before it reaches a shell.

No dedicated tool traced your code from untrusted OAuth/endpoint URL source → string interpolation → shell/browser-open sink, simulated whether the payload survives the quoting used, and checked your lockfile and MCP configs. This is that tool.

Setup

Requires Python 3.9+. No third-party dependencies — there is no requirements.txt because there is nothing to install.

git clone https://github.com/shouvikkkk/oauth-sink-check.git
cd oauth-sink-check
python -m oauth_sink_check scan ./your-project
# or, after pip install . :
oauth-sink-check scan ./your-project

Usage

oauth-sink-check scan <path> [--format text|json] [--strict]
  • scan accepts a file or a directory (directories are walked recursively, skipping node_modules, .git, __pycache__).
  • --format json emits a machine-readable report.
  • --strict makes LOW findings exit 1 as well.

Exit codes (exact scheme)

CodeMeaning
0Clean, or LOW findings only
1At least one MEDIUM finding, none higher (with --strict, LOW also exits 1)
2At least one CRITICAL or HIGH finding
3Usage error (missing/unreadable path, bad arguments)

Severities

SeverityMeaning
CRITICALTainted URL reaches a shell sink and a $(…) payload survives its quoting simulation
HIGHVulnerable open pin (CVE-2026-119025 range 7.2.0–<11.0.0), affected consumer version (opencode < 1.18.33, inspector < 2.12.0), or a literal $(…)/backtick payload in an MCP config
MEDIUMUnvalidated URL reaches a browser-open sink; tainted URL reaches a sink whose quoting neutralises the payload; remote MCP server with auth but no pinned client version info
LOWInformational — e.g. unpinned lockfile entry, open version predating the CVE range (unknown/legacy, not confirmed vulnerable), remote MCP server without auth

Example output

Scanning examples/vulnerable (exit code 2):

oauth-sink-check: examples/vulnerable

Findings: 1 CRITICAL, 1 HIGH

[CRITICAL] client.js:13 tainted-shell-sink
  Tainted authorization_endpoint URL reaches shell sink open and a
  $(...) payload survives its double-quoted PowerShell embedding
  ($(...) subexpression expands inside a double-quoted / unquoted
  PowerShell string).
  chain: authorization_endpoint (line 9) -> authUrl -> open sink (line 13)
  evidence: await open(authUrl);
[HIGH] package.json:0 vulnerable-dependency
  open 10.1.2 is in the CVE-2026-119025 vulnerable range 7.2.0-<11.0.0
  evidence: declared=^10.1.0 resolved=10.1.2 source=package-lock.json

Scanning examples/clean exits 0: it pins open 11.0.4 and allowlists https plus encodeURIComponent before open().

What it checks

ModuleWhat it does
sources.pyUntrusted URL sources: OAuth fields (authorization_endpoint, token_endpoint, issuer, redirect_uri, authorization_server, registration_endpoint, …), MCP discovery fields, config values. JS/TS + Python.
taint.pyIntra-file taint: sources → variables → template literals, f-strings, + concatenation, .format(), % formatting, simple same-file function args/returns. Reports source→sink chains with file:line.
sinks.pyJS/TS: open(), exec/execSync, spawn with shell:true, PowerShell invocations. Python: os.system, subprocess with shell=True, os.popen. webbrowser.open is a lower-severity browser sink, never conflated with shell sinks.
powershell.pyQuoting simulator: double-quoted, single-quoted, -EncodedCommand (base64-decoded, then re-checked), unquoted. Models the vulnerable open < 11 behaviour: URL unescaped inside a double-quoted PowerShell string → $(…) executes.
lockfiles.pypackage.json + package-lock.json (accurate), yarn.lock and pnpm-lock.yaml (best-effort regex). open < 11.0.0 flagged vulnerable only inside 7.2.0–<11.0.0; below 7.2.0 is unknown/legacy, not vulnerable. Small semver range resolver for ^, ~, >=, <, `
mcpconfig.py.mcp.json / mcp.json / opencode.json / .cursor/mcp.json: remote servers, literal $(…)/backtick payloads in authorization_endpoint/issuer, auth-without-pinned-client-version.
report.pyText + JSON reports, severity-gated exit codes.

Differentiation

ToolWhat it misses that this catches
npm audit / CVE feedsWhether your code actually routes an untrusted URL into a sink; your MCP configs
Generic SASTPowerShell quoting semantics — $(…) inside a valid https URL
Scheme allowlistsThe payload is a valid https URL; https-only does not stop it

Honest limitations

Download Tool

Most Popular

View all →

Discover the most used tools by our community.

  • 1

    OsintHunter

    letox4/osinthunter

    🎯 The ultimate OSINT toolkit: investigation graph, confidence scoring, alias families, map, timeline and reports. 49 modules, 400+ sites, no API keys required.

    1.0k
  • 2

    ghostlock-app

    yukonga/ghostlock-app

    GhostLock One-Tap Execution App (CVE-2026-43499)

    593
  • 3

    looking-glass

    anonghosty/looking-glass

    CSIRT platform that ingests Shadowserver reports into MongoDB and replays them as interactive maps, analysis, findings, advisories and executive dashboards.

    381
  • 4

    ARTEX

    baarai/artex

    LLM multi-agent autonomous penetration testing platform with planner/worker agents, asset and exploration graphs, MITM traffic recording, and vulnerability retesting.

    321
  • 5

    CyberStrikeAI

    ed1s0nz/cyberstrikeai

    The system of action for AI-native cybersecurity—where intent becomes governed execution, evidence becomes operational memory, and every operation improves the next.

    316
  • 6

    redforge

    cybersecurity91/redforge

    Browser-side red team toolkit with 102 offline tools for recon, payload generation, credential attacks, crypto, encoding and engagement reporting.

    305

Explore all tools

Browse our collection of tools

View all tools →