Claude-OSINT: eight Claude skills that turn the agent into a scoped external-recon operator

What Claude-OSINT is

Claude-OSINT is a library of skills for the Claude skills system. Each is a structured SKILL.md file that primes Claude with expert-level methodology for one part of the offensive-reconnaissance problem, so that, dropped into your Claude environment, the agent behaves like a senior recon analyst that knows the techniques, tooling, edge cases and escalation paths, and stays in scope. The README's figures put numbers on the depth: 8 skills, 100+ recon capabilities, 80 secret-regex patterns, more than 80 dorks, 9 read-only credential validators, 27 attack-path templates and roughly 10,000 lines of structured tradecraft.

It is built by Sachin Sharma, whose focus is GenAI security research and who also authored the related Claude-BugHunter bundle. The intended users are security professionals doing the recon phase of authorized engagements, and the README claims about 85 to 90% practitioner coverage for that phase, backed by a 56-prompt self-evaluation reported as 56/56 pass with zero fabrications. As with any such tool, the important word is authorized: this is offensive tradecraft meant for targets you have permission to assess.

The core pair: how to think and what to reach for

The library is anchored by two skills the README calls the recon backbone. osint-methodology is the how to think skill: strategic and procedural, covering asset-graph discipline, a severity rubric, time budgeting, identity-fabric mapping and deliverable templates, in about 515 lines. offensive-osint is the what to reach for skill: a tactical arsenal of probe paths, regexes, payloads, scoring rules, curl one-liners and tool URLs, and at roughly 4,555 lines it is the largest single file in the set.

This split is the sensible core of the design. Recon done well is a discipline (what to look at, in what order, how to score it) as much as a toolkit (which command finds a given exposure), and separating the two lets Claude reason about strategy and then reach for the right tactic. The offensive-osint skill also ships two stdlib-only scripts: secret_scan.py, an 80-pattern secret scanner, and h1_reference.py, a HackerOne disclosed-reports reference agent.

Six organisation-grade depth skills

Beyond the core pair, six skills lift the library from single-target recon to enterprise-scale attack-surface reasoning. org-attack-surface maps a legal entity to its owned footprint, using GLEIF org-trees, reverse-WHOIS, crt.sh organisation queries and org-first RIR netblock recall, and is explicitly discover-only. email-domain-security produces a composite spoofability verdict (making the point that an SPF -all alone is not spoof-proof and that DMARC governs) plus SPF supply-chain analysis.

The others quantify and monitor. exposure-risk-quantification applies a FAIR-style 0 to 100 and A to F risk score with a dollar-denominated loss model and a board one-pager, with honesty caps tied to ownership and proof. continuous-exposure-monitoring is a re-scan and diff loop with CTI and ransomware-leak-site chatter and finding-lifecycle discipline. cloud-saas-exposure covers bucket ownership-gated severity, offline AWS account-ID decoding and dependency-confusion checks. identity-provider-recon maps tenants and federation and includes a pre-auth user-enumeration oracle, with what the README calls a hard enumeration boundary. Those boundaries and the discover-only labels matter: they are how the skills encode staying in scope.

Installing and using it

Installation is simple because the skills are plain files. Clone the repository and drop the skill directories into your Claude skills folder:

git clone https://github.com/elementalsouls/Claude-OSINT.git

Each skill directory is self-contained, with its own SKILL.md and README.md, and once dropped into ~/.claude/skills/ Claude auto-triggers on relevant phrases rather than needing to be invoked by name. The repository's structure is laid out clearly, with the skills under skills/, docs under docs/, four end-to-end engagement walk-throughs under examples/, and the 56-prompt self-evaluation under tests/.

In use, you describe what you are testing and the relevant skill loads. The example walk-throughs, a quick recon, a bug-bounty workflow, identity-fabric mapping and secret hunting, are the fastest way to see the intended flow and are worth reading before a real engagement. Because the skills are Markdown methodology plus a couple of stdlib Python scripts, there is little to install and everything is inspectable, which is reassuring for a tool whose whole job is to run recon techniques.

Staying in scope, and why that is built in

The single most important thing about a tool like this is the boundary between authorized testing and misuse, and Claude-OSINT builds that boundary into its methodology rather than leaving it to the user's goodwill. Several skills are labelled discover-only, identity-provider-recon carries a hard enumeration boundary, and the risk-quantification skill demotes findings when ownership or proof is weak. The osint-methodology skill's severity rubric and deliverable templates are the professional framing that keeps recon evidence-based and in scope.

That design reflects the reality of the field: the same techniques that map an authorized client's attack surface would be abuse against a target you have no permission to assess. The README frames every use around authorized red-team and bug-bounty engagements, and that framing is not decoration; it is the condition under which the tool is appropriate to use at all. Anyone deploying it should have explicit scope and authorization first.

Licence, maturity and cautions

Claude-OSINT is MIT licensed, with a separate LICENSE-CONTENT file, which matters because much of the value is the curated tradecraft rather than code, and the two may carry different terms. The repository has no tagged releases; the last push was on 2026-08-30, and it ships a SECURITY.md, a CODE_OF_CONDUCT.md and a tests directory with the self-evaluation prompts. The self-reported 56/56 pass with zero fabrications is encouraging for a knowledge bundle whose main failure mode would be confidently inventing techniques or findings.

The cautions are the ones inherent to offensive tooling. The coverage and pass-rate numbers are the author's own, so validate them against your own engagements rather than taking them as guarantees. More importantly, this is capability that is only lawful and ethical within authorized scope; used outside it, the same skills constitute unauthorised reconnaissance. Treat Claude-OSINT as a force multiplier for a professional who already operates under proper authorization, and keep the discover-only and enumeration boundaries the skills define.

Editorial conclusion

Install Claude-OSINT if you do authorized red-team or bug-bounty recon and want Claude to carry the methodology, tooling and edge cases rather than reconstruct them each time. Drop the skill folders into ~/.claude/skills/ and describe your in-scope target in plain English. Use it only within authorized scope; it is offensive-recon tradecraft, not a tool for testing systems you do not have permission to assess.

Frequently asked questions

Official sources

1. elementalsouls/Claude-OSINT on GitHub
2. Issues
3. License: MIT
4. README