The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the bird CLI tool from a third-party Homebrew tap (steipete/tap/bird). This source is not listed as a trusted vendor, and the integrity of the downloaded binary is not verified within the skill's instructions.
[CREDENTIALS_UNSAFE]: The tool is designed to access browser cookies (e.g., Firefox and Chrome) to authenticate with X/Twitter. This involves reading sensitive session and credential data directly from the user's local browser profiles, which constitutes access to sensitive file paths.
[COMMAND_EXECUTION]: The skill executes various commands via the bird CLI on the host system to perform actions such as reading, searching, and posting tweets.
[PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection from external content retrieved from X/Twitter.
Ingestion points: Fetches untrusted data from X/Twitter via bird read, bird thread, and bird search commands.
Boundary markers: Absent; there are no delimiters or instructions provided to the agent to disregard instructions potentially contained within the retrieved tweets.
Capability inventory: The skill is capable of posting to the web via bird tweet and bird reply, which could be exploited to manipulate the account or exfiltrate information.
Sanitization: No evidence of sanitization or validation is present to ensure that fetched content does not influence the agent's behavior maliciously.

bird