The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (documents, code repositories, wikis) to generate operational manuals and SOPs. Ingested data could contain hidden instructions targeting the agent context.
Ingestion points: references/operations-manual.md identifies 'Source materials' such as document paths, code repositories, and wiki URLs as input.
Boundary markers: No specific boundary markers or 'ignore' instructions are defined to delimit processed content from system instructions.
Capability inventory: The skill utilizes mcp__sequential-thinking__sequentialthinking for complex reasoning and includes a post-processing chain involving external reviewer and humanizer skills.
Sanitization: The skill lacks explicit sanitization or filtering for the content extracted from external wikis or code comments.- [DATA_EXPOSURE]: The skill documentation mentions standard procedures for handling electronic approvals and internal regulations, but does not hardcode credentials or access sensitive system paths. It recommends using .env for secret management, which is a safe practice.- [REMOTE_CODE_EXECUTION]: While it refers to processing code repositories, it does so for analysis and documentation purposes rather than execution. It does not contain patterns for piping remote scripts to a shell.- [COMMAND_EXECUTION]: The skill mentions using a sequential thinking MCP tool if available, but does not provide arbitrary shell execution capabilities to the user.

collab-process