The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes user-stored notes from the doubaoya.com API, which constitutes an attack surface for indirect prompt injection if the notes contain malicious instructions.
Ingestion points: User content is ingested via the search, read, and ask commands in scripts/mera.mjs.
Boundary markers: The instructions in SKILL.md implement boundary controls by requiring the agent to use specific citation formats and designated answer_notice fields for responses.
Capability inventory: The skill uses the Node.js fetch API to communicate with vendor-controlled endpoints at doubaoya.com.
Sanitization: The markNoEvidence function in scripts/mera.mjs provides basic sanitization by identifying and replacing service-level English placeholders with localized warnings before passing data to the agent.
[COMMAND_EXECUTION]: The skill uses scripts/selfcheck.mjs to execute its own mera.mjs script via child_process.spawn. This is limited to local testing and does not involve executing untrusted or external commands.
[CREDENTIALS_SAFE]: The skill follows security best practices by reading the DOUBAOYA_API_KEY from environment variables rather than hardcoding secrets. The scripts include explicit warnings to agents not to echo or log the API key.

mera