The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external codebases and processes metadata such as symbol names, which creates a potential surface for indirect prompt injection if the analyzed code contains adversarial symbol names.
Ingestion points: Codebase structure and symbol metadata are ingested into the agent context via the output of the claudemem command (SKILL.md).
Boundary markers: Absent. The skill does not employ specific delimiters or instruction-based guards to isolate data produced by the analysis tool from the agent's logic.
Capability inventory: The skill leverages Bash for command execution and data processing, and Read for file system access, which are exploitable if injected instructions are inadvertently followed.
Sanitization: Absent. The skill pipes raw output from the analysis tool directly into shell utilities like awk and echo without escaping or validating the content of symbol names or paths.

architect-detective