The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration logic in fact_check_x.py and trusted_search_config.py uses the subprocess module to execute sibling fact-checking engines, rendering tools, and local tests.
[EXTERNAL_DOWNLOADS]: The prepare-runtime function utilizes npm ci with the --ignore-scripts flag to install required dependencies (@playwright/test, commander, zod) from the public Node package registry.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted LLM outputs from results.json and interpolates them into search queries and reports. This attack surface is mitigated by the implementation of human review checkpoints (stage-checkpoints.json) and the use of html.escape for rendering technical notices.
[SAFE]: The management of the TRUSTED_SEARCH_KEY includes significant security safeguards, such as enforcing 0o600 Unix permissions for local storage, redacting the key from all console outputs via the redact_sensitive_json utility, and utilizing the RejectCredentialRedirects class to prevent credential exposure via HTTP redirects.

fact-check-x-unified