The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code during its discovery and research phases, creating a potential surface for indirect prompt injection. However, the skill implements strong mitigations:
Ingestion points: Files are read using localGetFileContent and Read tools across multiple agents.
Boundary markers: The Research Agent explicitly wraps analyzed code content in <code_content> delimiters.
Capability inventory: The skill uses Task and Task_Parallel for orchestration and Write for file generation.
Sanitization: The Research Agent instructions include a mandatory content_boundary_protocol that forbids following any instructions found within code comments or string literals, treating them strictly as evidence.
[REMOTE_CODE_EXECUTION]: The documentation provides instructions for setting up the octocode-mcp server using npx. This is a standard distribution mechanism for Node.js-based MCP servers and originates from the skill's own author (bgauryy).
[COMMAND_EXECUTION]: The skill makes extensive use of the Task and Task_Parallel tools to orchestrate its 6-phase pipeline. These calls are well-structured, use specific model hints, and include built-in retry and state preservation logic to ensure reliable execution without bypassing safety constraints.
[DATA_EXPOSURE]: The skill inherently requires access to the repository's source code to perform its primary function of generating documentation. It uses scoped tools like localSearchCode and localViewStructure to minimize unnecessary file access.

octocode-documentation-writer