The Agent Skills Directory

[PROMPT_INJECTION]: The skill's primary function is to process and analyze untrusted output from external AI platforms (e.g., DeepSeek, Doubao). This creates an inherent surface for indirect prompt injection (Category 8).
Ingestion points: External AI responses are collected via browser automation and stored in capture/results.json for analysis.
Boundary markers: The instructions enforce a multi-stage workflow in SKILL.md that separates raw data collection from comparison and authoritative verification steps.
Capability inventory: The skill infrastructure includes capabilities for local command execution (via testing scripts) and network access for verification APIs.
Sanitization: The skill utilizes atomic fact splitting and alignment logic to normalize claims before verification, reducing the likelihood of malicious instructions influencing the final report.
[COMMAND_EXECUTION]: Multiple scripts within the package, such as golden_regression_gate.py, workbuddy_acceptance.py, and run_golden_case.py, utilize the subprocess module to execute local Python and Node.js components. These operations are used for regression testing, integrity validation, and orchestrating the internal fact-checking pipeline.
[DATA_EXFILTRATION]: The skill manages verification API keys by storing them in a local user directory (~/.fact-check-x/). The logic in release_gate_core.py and SKILL.md includes explicit security measures to prevent these credentials from being leaked in logs, chat responses, or generated reports. Network communication is targeted towards specific AI platform domains and official verification endpoints.

fact-check-x-complete