browser-automation

SUSPICIOUS: the core browser automation capability matches the stated purpose and the main CLI install path is official, but the skill is high-impact. It teaches token extraction, request capture, and authenticated internal-API use, and it enables autonomous public posting/messaging and job-application actions. The biggest concerns are credential exposure through browser-state handling and the mutable transitive update path via `npx skills update`, not confirmed malware.

No explicit classic malware indicators (obfuscation, credential theft, persistence, or system/network exploitation code) are visible in the provided fragment. However, the described functionality is highly privacy-sensitive: it enables harvesting WhatsApp message content (including large history via scrolling), automating outbound message/media sending, and extracting/transcribing voice-note audio from browser Cache Storage with subsequent off-browser processing. This combination aligns with surveillance/data-exfiltration tooling if used without clear user authorization and strong guardrails around data handling and external transmission.

This module is not a traditional malware payload, but it is a high-risk traffic capture utility that can record sensitive authenticated request data (headers and POST bodies, potentially including authorization/CSRF tokens and private chat/message contents) and persist session state to disk. Its broad filtering and lack of redaction make accidental or misuse disclosure likely; logs and the persistent profile should be treated as sensitive and tightly controlled to authorized, permitted use only.

No concrete evidence of classic malware (backdoor, persistence, or payload execution) is present in this fragment. However, it substantially enables misuse: it teaches reverse-engineering by capturing and logging full request headers/bodies (which can include authentication tokens) and it instructs extracting tokens from localStorage to perform authenticated, state-changing chat operations including soft-delete semantics. Treat as a moderate-to-high misuse/credential-exposure risk when used outside strict authorization and logging controls.

No strong indicators of embedded malware (no obfuscated payloads, no exfiltration to external domains, no backdoor mechanics) are present in this fragment. However, it is explicitly an automation-focused reference that enables authenticated, state-changing actions on a user’s LinkedIn session—extracting CSRF from cookies and scraping opaque identifiers from DOM HTML to call undocumented messaging and attachment-upload endpoints. The dominant risk is misuse for account abuse/spam/unauthorized outreach capability rather than overt malicious code in this snippet.

The provided fragment is not overtly malicious in the classic malware sense (no obfuscation or exploit-style primitives are evident), but it is explicitly designed for harvesting personal owner contact information from third-party property directory pages and supports bulk automated WhatsApp outreach using the harvested phone numbers. This creates a high likelihood of privacy/consent violations and spam/abuse risk, making the package/project high-risk from an abuse-prevention and policy-compliance perspective. Additional code (e.g., the actual WhatsApp automation implementation and any file logging) would be needed to confirm the exact behavior beyond what is described.

This content provides a highly actionable technique for harvesting authentication/session tokens from browser `localStorage` via dynamic `eval`, exporting them to the caller, decoding JWT claims, and refreshing tokens (e.g., Cognito `refresh_token` flow) to sustain unauthorized access. The behavior aligns strongly with credential theft/session hijacking rather than legitimate automation or testing. Obfuscation is minimal; risk stems from direct secret extraction and abuse enablement.