The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from GitHub issue titles, descriptions, comments, and pull request diffs or metadata to draft documentation. This creates an attack surface where malicious instructions embedded in these external sources could influence agent behavior. \n
Ingestion points: Data is retrieved from external sources using gh issue view, gh pr view, gh pr diff, and WebFetch operations. \n
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded commands within the processed issue and PR content. \n
Capability inventory: The agent possesses significant capabilities, including file modification (Edit, Write) and repository management (Bash with gh and git commands), which are used to create branches and open pull requests. \n
Sanitization: There is no documented process for sanitizing or filtering external content before it is incorporated into prompts or used for drafting. \n- [DYNAMIC_EXECUTION]: The skill executes external code via npx as a diagnostic measure. \n
Evidence: The command npx @modelcontextprotocol/inspector --url https://www.elastic.co/docs/_mcp/ is used to check the reachability of the documentation service's Model Context Protocol (MCP) server. \n
Context: While used for diagnostic purposes, this involves the dynamic download and execution of a Node.js package. \n- [EXTERNAL_DOWNLOADS]: The skill fetches documentation guidelines and configuration from external repositories and domains. \n
Evidence: The skill references and downloads content from official vendor domains such as elastic.co and vendor-owned GitHub repositories under the elastic organization. \n
Context: These resources are official infrastructure for the skill's authoring organization and are documented here as standard operational behavior. \n- [COMMAND_EXECUTION]: The skill performs shell commands to manage version control and documentation workflows. \n
Evidence: The skill uses git for branch management and gh (GitHub CLI) for interacting with issues and pull requests. \n
Context: These operations are restricted to specific, non-administrative tools and are governed by explicit human approval gates before critical actions like file writing or PR submission occur.

docs-draft-feature-docs