The Agent Skills Directory

Data Exposure & Exfiltration (HIGH): Accesses sensitive browser data and authentication tokens. The tool is designed to extract cookies directly from browser profile directories (e.g., via --chrome-profile-dir and --firefox-profile). This provides the agent with access to sensitive credential databases that may contain more than just X/Twitter authentication tokens.
Unverifiable Dependencies & Remote Code Execution (HIGH): Installs software from untrusted sources and uses remote execution. The skill installs @steipete/bird via NPM and steipete/tap/bird via Homebrew. These sources/authors are not on the trusted whitelist. The instruction bunx @steipete/bird whoami downloads and executes code from an external registry at runtime without pre-verification.
Indirect Prompt Injection (LOW): Vulnerable to instructions embedded in external social media content.
Ingestion points: The skill reads tweets, threads, and search results via bird read, bird thread, and bird search.
Boundary markers: None are specified to protect the agent from instructions embedded in tweets.
Capability inventory: The skill has powerful capabilities including posting tweets (bird tweet) and managing social connections (bird follow).
Sanitization: No evidence of sanitization for the ingested content is provided.
Command Execution (LOW): Executes shell commands to interact with the bird CLI as part of its primary functionality.

bird