The Agent Skills Directory

[DATA_EXFILTRATION]: The skill processes user-provided LinkedIn connection exports (Connections.csv and Profile.csv). Relevant identifiers are sent to the author's (crustdata) enrichment API for relationship mapping. This data access is transparently documented and serves the skill's primary purpose via the author's own infrastructure.
[COMMAND_EXECUTION]: The skill invokes bundled Python scripts (parse_connections.py and build_workspace.py) to process data and generate an interactive HTML report. These scripts operate locally on the data provided and do not initiate independent network connections.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external social media posts and CSV files, creating a surface for indirect injection. This is mitigated by the following:
Ingestion points: LinkedIn connection files and social posts are analyzed via a sub-agent (references/stakeholder-agent.md).
Boundary markers: The sub-agent is strictly constrained to return only a JSON object and use only specific data tools.
Capability inventory: The skill uses execute for local processing, callTool for enrichment, and Artifact to publish results.
Sanitization: The agent is instructed to focus on structured data extraction rather than open-ended natural language interpretation of external content.

warm-path-workspace