The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection (!```bash) in SKILL.md to process user-supplied input through the $ARGUMENTS variable.
Evidence: The script captures $ARGUMENTS using a heredoc (<<'__MEMADD_EOF__'). If the user provides input containing the delimiter __MEMADD_EOF__, they can terminate the heredoc prematurely and inject arbitrary shell commands that will execute when the skill is invoked.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by persisting user data into a shared memory file (Category 8).
Ingestion points: User input from $ARGUMENTS is appended to ~/.claude/memory/staging.md by the append.sh script.
Boundary markers: Absent. Content is only prefixed with a bullet point (- ), which does not provide a clear boundary to prevent the agent from misinterpreting stored text as instructions during future retrieval.
Capability inventory: The skill possesses file system write capabilities (mkdir, printf >>).
Sanitization: The append.sh script validates that input is not empty and is a single line, but it does not sanitize the content for nested instructions or command injection sequences.

memory-add