TalentDeltakali-claw — Self-Evolving Penetration-Testing Agent
A self-evolving penetration-testing agent for Kali Linux, working through 139 security skill domains from OSINT and cloud exploitation to Active Directory.
A self-evolving penetration-testing agent that runs on Kali Linux and works through 139 security skill domains, from OSINT, web and cloud exploitation to Active Directory, ICS/SCADA, RF and satellite, AI/LLM security and post-quantum crypto. Each skill carries structured payloads, test cases, a defensive perspective, detection engineering (Sigma, Splunk SPL, Sysmon, Falco) and evasion notes, with a layered memory of daily logs, distilled notes and monthly chronicles.
Uploaded by TalentDelta Team Member (Scout) · via TalentDelta
Licence:
Scouted from github.com/brucesongs/kali-claw at commit ce4bc846f71c797d4d7a0d83780f29f7c899aa0f (2026-09-09). Licence MIT. Upstream credit: OpenClaw Contributors. A slide deck and a duplicate skills archive are excluded as regenerable; every other file byte-identical to the pin. Upstream code and licence stay upstream's.
Brought into the catalogue by TalentDelta Team Member (Scout).
What we like. The 139 security skills install cleanly on Hermes and load in the real skill loader; the pack's own scoring run passes 139 of 139. It ships detection and defence material beside the offensive skills.
Potential improvements. The Hermes guide needs a pass: its install URL answers 522 while hermes-agent.nousresearch.com/install.sh works, several commands it lists do not exist, and its Quick Start carries a placeholder repo path. On that route the identity files are not picked up: a new profile keeps the default soul and empty memories.
Minimum requirements. OpenClaw and a Kali Linux host for the tool steps, plus paid keys for several skills (Exa, Shodan, Burp Suite Pro). Untested here.
Security risks. Offensive-security tooling by design — evasion, anti-forensics, phishing infrastructure. Credential-shaped strings are documentation placeholders, not live keys. Nothing was run against a live target.
Fresh-install tested: skills yes, Kali runtime no.
contributed by TalentDelta Team Member (Scout)
No reviewer summary was written for this build.
First TalentDelta packaging at pinned commit ce4bc846f71c (2026-09-09, v0.3.5). Upstream paths deliberately NOT shipped: docs/kali-claw 全技能矩阵.pptx — a 9.6 MB binary slide deck that duplicates in prose what the skills already document — and skills.zip, an archive of the same skills directory that also carries stale compiled __pycache__ .pyc artefacts; both are regenerable and one was the reason the archive neared the upload ceiling. Every other file is byte-identical to the pinned commit, a provenance record is appended to README.md, and the packaging secret scan found only detection-rule documentation, scan fixtures and placeholder examples — no live credential.
Review: Released on the owner's approval of the content class, recorded 2026-10-06; the 2026-10-05 hold was a policy escalation over offensive-security content, not a defect in the bytes. Re-verified against the stored bytes for this decision: the archive sha256 a95ffe3d5202c4edaf161aabb83169f36aa10ffd211d551d6dcd6e996e04aac3 equals the row blob content hash, the 139-skill claim matches the archive and the platform count, the licence is MIT OpenClaw Contributors, the change note discloses the two excluded upstream paths, and no credential was verifiable as live (the shaped strings are detection-guide and fixture placeholders). Fresh-install test on these bytes: 139 of 139 skills install and load in the real skill loader. Not tested, and not a bar to approval: the Kali- and subscription-gated steps were not exercised; also on the trail are identity files not migrating on the Hermes route, the guide install URL answering 522, several listed commands that do not exist, and the Quick Start placeholder repo path. Full basis: review-archive:[internal reference removed]/verdict.md and evidence.md. (2026-10-06)