The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements proactive input sanitization by instructing the agent to truncate arguments to 200 characters and strip dangerous characters such as backticks and angle brackets. It also employs boundary markers when passing user data to sub-agents, which helps prevent the interpretation of research data as instructions.
[COMMAND_EXECUTION]: The skill uses Bash to perform state updates via jq. It correctly utilizes safe parameter passing by using jq's --arg and --argjson flags, which ensures that variable content is treated as data rather than part of the executable filter, effectively preventing command injection.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its orchestration of sub-agents that perform web research. 1. Ingestion points: Data retrieved from the web via sub-agents and read from findings files. 2. Boundary markers: The lead agent uses boundary tags when defining tasks for sub-agents. 3. Capability inventory: The skill has Bash shell, filesystem write, and agent spawning capabilities. 4. Sanitization: The skill enforces a 'Structured Data Protocol' which requires schema validation after every file modification to maintain data integrity and filter unexpected structures.
[DATA_EXFILTRATION]: No unauthorized data exfiltration patterns were identified. The skill uses standard internal messaging and local file storage for its research data pipeline.
[EXTERNAL_DOWNLOADS]: Tools for web research and diagram generation are used, which are appropriate for the skill's intended purpose and originate from well-known technology services.

augment