The Agent Skills Directory

[PROMPT_INJECTION]: The skill generates instructions for the agent to explicitly ignore platform-level features, specifically the built-in auto-memory system (~/.claude/auto-memory). This is an instruction override pattern that directs the agent to disregard standard system behavior in favor of the skill's local configuration.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an automated system that reads from and writes to local markdown files at the start and end of every session, which can be exploited to influence agent behavior through data persistence.
Ingestion points: The agent is instructed to read MEMORY.md, memory/YYYY-MM-DD.md, and scan project files (README, source code) to define its role and context.
Boundary markers: The generated CLAUDE.md and MEMORY.md templates do not contain delimiters or 'ignore embedded instructions' warnings for the data stored within the memory files.
Capability inventory: The skill utilizes shell commands (mkdir) and instructs the agent to use Read and Grep tools to interact with the local filesystem.
Sanitization: No sanitization or validation is performed on the data being written to or read from the memory files, allowing potentially malicious instructions in processed project data to become part of the agent's long-term personality and rule set.
[COMMAND_EXECUTION]: The skill uses a shell command (mkdir -p memory/) to prepare the filesystem for its operation. While functional for the stated purpose, it demonstrates the ability to execute arbitrary commands during the setup phase.

memory-init