The Agent Skills Directory

[PROMPT_INJECTION]: No malicious injection or behavior override patterns were detected. The instructions prioritize evidence-based reasoning and neutral reporting.\n- [DATA_EXFILTRATION]: No unauthorized data access or exfiltration patterns were found. The skill is designed to process information within established source boundaries.\n- [REMOTE_CODE_EXECUTION]: No remote code execution, package installation, or dynamic code execution patterns were identified.\n- [INDIRECT_PROMPT_INJECTION]: The skill manages potential risks from external data ingestion through a structured evidence validation framework. 1. Ingestion points: External policy documents, platform rules, and media reports referenced in SKILL.md. 2. Boundary markers: The skill implements a 'source ledger' and 'evidence ladder' to explicitly define the reliability of ingested data. 3. Capability inventory: Uses standard reasoning tools (Claude, Codex) without file-write or subprocess capabilities. 4. Sanitization: Instructions mandate identifying what sources 'cannot prove' to prevent the accidental adoption of instructions from external content.

crossframe-public