The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of the nmem CLI utility for its retrieval logic. Commands such as nmem --json m search "query" and nmem --json t show <thread_id> use string interpolation to build shell commands. If the agent does not properly sanitize input variables like the search query or thread identifier, it could potentially lead to command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the retrieval and ingestion of potentially untrusted data from an external memory store into the agent's context, which is a known attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through the results of memory_search, thread_search, and nmem search/show commands described in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters used to separate retrieved content from agent instructions or to warn the agent about ignoring instructions found within the memories.
Capability inventory: The skill environment includes the ability to execute shell commands (via nmem) and perform graph-based exploration (via explore-graph).
Sanitization: The instructions do not specify any sanitization, escaping, or validation of the data retrieved from the knowledge base before it is processed or presented.

search-memory