In the wild
Typosquatted agent skills and trojanized PyPI packages deliver a second-stage credential harvester
AIR-2026-0188 AIR record · 2026-08-zenity-skill-supply-chain-campaign
From the supporting source
The skills were first weaponized when malicious setup-installation.md was added in seven skill paths, instructing agents to fetch and execute the remote payload.labs.zenity.io
The record
Disclosed
2026-08-06
Occurred
2026-08-02
Vendor
skills.sh / PyPI
Product
skills.sh skills getpaperclipai/paperclip and browser-use-headless/browser-use-headless-skill; PyPI packages paperclip-ai 0.1.0/0.1.1 and browser-use-headless 0.1.4
Risk domain
Tool misuse › Supply chain vulnerabilities
Also touches
Credential theft
System type
Generative model
Who acted
Agent acted — the agents installed the typosquatted skills and fetched the second stage
Attack surface
Skill / plugin
Autonomy level
Semi-autonomous
Initial vector
Malicious marketplace package
Impact type
Credential theft, Malware installation
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI04 – Agentic Supply Chain Vulnerabilities
- MITRE ATLAS
- AML.T0040 – AI Supply Chain Compromise
- NIST AI RMF
- MAP 4.1
- EU AI Act
- Annex III §2
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool failure
- Mitigation
- Scan dependencies and container images for vulnerabilities; Use trusted sources.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
More than 1.7M aggregate installs across the skill family by 2026-08-02
Capability abused
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
Campaign disrupted after 2026-08-02; Zenity credits Vercel and Microsoft/GitHub for quick response
Taxonomy & references
Curator notes
Author Michael Bargury (Zenity Labs). Install counts are aggregate across the skill family, not confirmed compromises, so impact_realized is left false; Zenity advises treating any agent host as compromised if the loader or stealer executed.
Cite as
Enkrypt AI. "Typosquatted agent skills and trojanized PyPI packages deliver a second-stage credential harvester". AIR, AIR-2026-0188. https://agentincident.vercel.app/i/AIR-2026-0188/ (accessed 2026-10-19).