The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download external content.
It recommends cloning a third-party repository to access the full analysis series: https://github.com/luyao618/Claude-Code-Source-Study.
It also references the official Claude Code repository from Anthropic: https://github.com/anthropics/claude-code.
[INDIRECT_PROMPT_INJECTION]: The skill provides code patterns for an agent scaffold that processes untrusted data, which introduces an attack surface for indirect prompt injection.
Ingestion points: The userInput variable is used in the Minimal Agent Scaffold section within SKILL.md.
Boundary markers: The provided code snippets do not implement specific delimiters or instructions for the agent to ignore commands embedded within the user input.
Capability inventory: The demonstrated capabilities include file system access via fs.readFile and message creation via the Anthropic API, as shown in SKILL.md.
Sanitization: The example code does not include validation, escaping, or filtering of the user-provided input before it is sent to the model or used in file operations.

claude-code-source-study