The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to publish local project patterns to the public IPFS network using a Pinata JWT. While the documentation provides a warning about stripping PII (Personally Identifiable Information) before publishing, the core function involves sending project-derived data to an external, publicly accessible storage system.
[INDIRECT_PROMPT_INJECTION]: The skill ingests patterns from external IPFS Content Identifiers (CIDs) through the 'load' action.
Ingestion points: Patterns are fetched from remote IPFS CIDs and applied to the local project environment via the hooks_transfer tool (SKILL.md).
Boundary markers: No boundary markers or instruction-ignore delimiters are documented for the ingested pattern content.
Capability inventory: The skill utilizes tools for pattern storage, search, and general shell access (Bash), meaning ingested patterns could potentially influence subsequent tool selection or logic.
Sanitization: There is no evidence of sanitization, validation, or structural checking for patterns fetched from external peers.
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands and MCP tool calls to interact with the local file system and network, specifically mirroring patterns from arbitrary source paths on the local disk using the from-project action.

intelligence-transfer