The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified through repository-specific documentation.
Ingestion points: The skill instructions in both SKILL.md and README.md specify that if the AGENTS.md file exists in the repository root, it will be added to the agent's context automatically.
Boundary markers: Absent. There are no specific instructions or delimiters defined to prevent the agent from obeying instructions that might be embedded within the AGENTS.md file.
Capability inventory: The skill facilitates file system writes and modifications specifically for the AGENTS.md file.
Sanitization: Present. The skill instructions mandate that the agent must summarize proposed information and obtain explicit user confirmation before writing any data to the repository, which serves as a significant manual checkpoint.

agent-memory