The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to access and utilize sensitive authentication credentials (specifically 'auth_token' and 'ct0' cookies) from the local storage of major web browsers including Chrome, Firefox, Edge, and Safari. This data is used to authenticate requests to the Twitter/X API via a remote proxy or local MCP process.
[COMMAND_EXECUTION]: The installation and setup instructions require the execution of code from the NPM registry using 'npx @checkra1n/xbird' or 'bunx @checkra1n/xbird'. This is a standard distribution method for the tool but involves running third-party code on the host machine.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it ingests untrusted data from the Twitter platform which could contain malicious instructions for the AI agent.
Ingestion points: Data enters the agent context through tools such as 'get_tweet', 'get_replies', 'get_home_timeline', and 'search_tweets' as defined in tools.md.
Boundary markers: No explicit delimiters or instruction-ignore warnings are specified in the SKILL.md files to separate untrusted Twitter content from system instructions.
Capability inventory: The agent has significant write capabilities including 'post_tweet', 'update_profile', 'follow_user', and 'upload_media', which could be abused if an injection is successful.
Sanitization: No evidence of content sanitization or instruction filtering is present in the provided skill documentation.

xbird