The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to construct and execute shell commands using potentially untrusted user input. Specifically, the template claudemem search "[query]" takes a user-supplied topic and interpolates it into a shell command. If a user provides input containing shell metacharacters (e.g., semicolons, backticks), it could lead to unintended command execution within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user data to drive code investigation tasks, creating a surface for indirect prompt injection.
Ingestion points: User queries containing keywords such as "audit", "investigate", "how does X work", and "trace" are ingested through the skill's auto-trigger logic in SKILL.md.
Boundary markers: No boundary markers, delimiters, or "ignore embedded instructions" warnings are utilized when interpolating these queries into search commands or subagent tasks.
Capability inventory: The skill employs the Bash tool for executing claudemem and grep commands, and the Read tool for accessing file contents across the codebase.
Sanitization: The instructions do not provide any sanitization, escaping, or validation logic to filter user-supplied search terms before they are used to influence agent behavior or shell execution.

code-search-selector