← MCP server and skill grades

skills

UseOSINT/Skills/skills

B
Type
Skill
Scanned
Nov 13, 2026
Version
06243a5620b0
Files read
80

This project is a collection of Markdown guides that teach an AI agent how to conduct open-source intelligence (OSINT) research on domains, companies, people, accounts, media files, and code repositories. Its main risks are that the default router skill tells the agent to fetch and trust live instructions from useosint.com over the locally installed files, and the git-history skill runs TruffleHog with live credential verification left on by default, which can make active login attempts and network connections when scanning untrusted repositories.

What it can reach

Network
Instructs the agent to query external OSINT services and websites including useosint.com, crt.sh, rdap.org, Shodan, Censys, GitHub API, Wayback Machine, Common Crawl, archive.today, and reverse-image search engines.
Shell commands
Provides command recipes for the agent to run local CLI tools such as curl, wget, dig, whois, git, trufflehog, gitleaks, exiftool, ffmpeg, ffprobe, magick, tesseract, yt-dlp, subfinder, amass, dnsx, shodan, sherlock, maigret, phoneinfoga, unzip, pdfimages, and qpdf.
Files
Instructs the agent to read, copy, hash, extract, and save local media files, Office and PDF documents, git mirror clones, WARC web captures, and CSV or Markdown investigation reports.
Browser
Guides browser-based searches across search engines, public social and messaging previews, block explorers, and web archives when browser tooling is available to the agent.

Findings

2 medium1 low
  1. MediumRouter skill instructs the agent to fetch and trust unpinned live instructions from useosint.com

    skills/useosint/SKILL.md:31

    The useosint entry-point skill is enabled for automatic model invocation on broad everyday prompts such as 'research', 'verify', 'look up', or 'who is this'. It instructs the agent to prefer live retrieval from useosint.com (including catalog.json and remote .md skill files) over local knowledge and to trust the live documentation whenever they disagree. If useosint.com is ever compromised, changes ownership, or serves modified content, new instructions can be injected into the agent without a skill update. Fetching remote .md workflows also sends usage telemetry to useosint.com and lets the agent bypass the disable-model-invocation safety gates placed on ten sensitive local skills such as find-anyone and hunt-a-handle.

    Fix: Remove the instruction to fetch live remote documentation over local references, rely solely on the locally installed Markdown files, and add disable-model-invocation: true to useosint or avoid fetching gated skill files over HTTPS.

  2. MediumTruffleHog commands run live credential verification and network probes by default

    skills/secrets-in-git-history/SKILL.md:133

    Although this skill warns never to authenticate with a discovered credential and advises against verification outside authorized assessments, the trufflehog commands in both SKILL.md and reference/git-archaeology.md omit the --no-verification flag. In TruffleHog v3, active credential verification is enabled by default (--results=verified only filters displayed output). Running the command as written against an untrusted repository causes TruffleHog to actively test candidate secrets against third-party APIs and open outbound connections to database, webhook, or service URLs embedded in commit history, which can trip attacker canary tokens, expose the user's IP address,probe internal hosts, and violate computer-misuse boundaries.

    Fix: Append --no-verification to all default trufflehog commands in skills/secrets-in-git-history/SKILL.md and skills/secrets-in-git-history/reference/git-archaeology.md, and explicitly note that TruffleHog performs live verification unless --no-verification is supplied.

  3. LowWayback Machine CDX API queries use unencrypted HTTP

    skills/read-deleted-pages/SKILL.md:70

    The curl commands for querying the Wayback Machine CDX API in skills/read-deleted-pages/SKILL.md, skills/read-deleted-pages/reference/cdx-cookbook.md, and skills/find-hidden-subdomains/reference/subdomain-sources.md use cleartext http://web.archive.org instead of https://web.archive.org. Anyone on the local network can observe which domains and paths are being investigated and could tamper with the unencrypted JSON responses fed back into the agent's context.

    Fix: Replace http://web.archive.org/cdx/search/cdx with https://web.archive.org/cdx/search/cdx across all curl examples.