The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands by directly interpolating user-controlled variables (such as target names and technique descriptions) into a command line executed via uv run.
Evidence: Multiple entries in SKILL.md utilize this pattern, notably: uv run python3 $CLAUDE_PROJECT_DIR/tools/brain.py record <target> <status> <technique> "<details>".
Risk: If input variables contain shell metacharacters like semicolons or pipes, they could lead to execution of arbitrary commands.
[PROMPT_INJECTION]: The skill is designed to ingest and store untrusted data from external 'targets' (endpoints, files, workflows) into a memory system (MEMORY.md), creating an indirect injection surface.
Ingestion points: User-supplied inputs for target, status, technique, and details fields in SKILL.md.
Boundary markers: Quotes are used for the <details> field in the command execution, but no explicit boundary markers or instructions to ignore embedded commands are present for the other fields.
Capability inventory: The skill executes local Python scripts via uv run as documented in SKILL.md.
Sanitization: There is no evidence of sanitization, validation, or escaping logic applied to the input variables before they are passed to the shell or stored in memory.

brain