The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local bash script (validate-enriched-trends.sh) to validate and repair JSON data produced by subagents and performs routine file cleanup within the project directory. These operations are limited to project-specific paths and follow standard development practices.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it processes web-sourced evidence through subagents. This is an inherent risk for research skills, which is mitigated by output validation gates.
Ingestion points: The skill reads project data from trend-scout-output.json and web-sourced data processed by trend-report-writer agents.
Boundary markers: The orchestrator's dispatch prompts do not explicitly use delimiters to isolate external content.
Capability inventory: The skill can execute local bash scripts, dispatch parallel subagents, and write metadata manifests.
Sanitization: Employs a specific script to validate the JSON structure of subagent output and repair common LLM formatting errors.
[SAFE]: No sensitive data exposure, hardcoded credentials, or unauthorized network operations were identified. The skill's behavior aligns with its documented purpose and utilizes vendor-owned subagent resources.

trend-research