The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands (grep, sed, pwd) to manage project-specific memory files. The <keywords> placeholder in the duplicate check step (grep -ni "<keywords>" "$MEMORY_DIR/MEMORY.md") is derived from user-provided input. If not properly escaped by the agent, this could allow for command injection (e.g., using backticks or subshells within the keywords).
[INDIRECT_PROMPT_INJECTION]: The primary function of the skill is to write user-supplied text to MEMORY.md, a file intended to be read by the agent at the start of every session. This creates a persistence mechanism for potential malicious instructions.
Ingestion points: User-provided string processed in 'Step 1: Parse the knowledge' of the SKILL.md workflow.
Boundary markers: Minimal; instructions suggest formatting as a markdown list item (- {{concise fact or pattern}}).
Capability inventory: The skill has capabilities to read files (via grep) and write/append to the filesystem (MEMORY.md).
Sanitization: The skill includes a 'What NOT to use' section warning against storing sensitive data, but lacks functional sanitization or escaping of the input content before it is written to the persistent memory file or used in the shell command.

remember