The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests natural language input and writes it directly into files that govern agent behavior.
Ingestion points: The thing to remember argument in SKILL.md is used to capture user-provided rules.
Boundary markers: None. The instructions do not specify a requirement to wrap the user-provided content in delimiters or ignore potentially malicious embedded instructions.
Capability inventory: The skill uses the Write and Edit tools to modify documentation (AGENTS.md, CLAUDE.md) and skill definitions (SKILL.md).
Sanitization: None. The skill does not validate or sanitize the content before persisting it to the repository's knowledge base.
[COMMAND_EXECUTION]: The skill leverages the Bash tool to manage repository structure, specifically for creating directories and symlinks as described in the conflict resolution and synchronization steps.
[DYNAMIC_EXECUTION]: The skill generates and modifies SKILL.md files at runtime to create repo-specific overrides for shared skills. These files contain executable instructions that the agent will follow in future sessions.

remember