The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The package.json and bun.lock files specify version numbers for @anthropic-ai/sdk (0.78.0) and playwright (1.58.2) that do not currently exist in the official NPM registry. This indicates a potential supply chain attack or the use of malicious dependencies from an unverified mirror.
[DATA_EXFILTRATION]: The src/cookie-import-browser.ts file implements functionality to programmatically access the macOS Keychain via the security command to retrieve storage passwords and decrypt session cookies from Chrome, Edge, and other local browsers. This provides a direct path for the agent to harvest and potentially exfiltrate sensitive authentication tokens.
[REMOTE_CODE_EXECUTION]: The ubuntu-cmd-install.txt file contains instructions to add an unverified third-party repository (debian.griffo.io) and trust its GPG keys, which involves executing remote code and trusting software from an unvetted source.
[COMMAND_EXECUTION]: The skill instructions in ubuntu-cmd-install.txt include the use of sudo to perform system-level modifications, representing a privilege escalation risk.
[EXTERNAL_DOWNLOADS]: Setup instructions in SKILL.md guide the download and execution of the Bun installer from https://bun.sh/install using a shell pipe.
[COMMAND_EXECUTION]: The preamble in SKILL.md automatically executes bash commands to check for updates and manage local session state using binaries located in the user's home directory.

sdlc-qa-browse