The Agent Skills Directory

[COMMAND_EXECUTION]: The integrity check in Step 6 uses a shell for loop to iterate over links extracted from MEMORY.md using grep. The extracted filenames are interpolated directly into an ls command without quoting or sanitization. If MEMORY.md contains a malicious link with shell metacharacters (e.g., [test](; rm -rf / ;)), the command could execute arbitrary shell code.
[PROMPT_INJECTION]: Indirect Prompt Injection surface detected (Category 8).
Ingestion points: Content is read from MEMORY.md and other topic files in the memory directory (SKILL.md, Step 6).
Boundary markers: None. Content extracted via regex is treated as trusted input for shell execution.
Capability inventory: Uses Bash for integrity checks and Write/Edit for file management.
Sanitization: Absent. Filenames extracted from markdown links are passed directly to the shell environment.

manage-memory