The Agent Skills Directory

[COMMAND_EXECUTION]: The skill-run script executes files based on a path provided as a command-line argument. It constructs the target path by prepending a skills root directory but fails to validate that the resulting path remains within that directory. This allows for path traversal (e.g., using ../../ sequences) to execute arbitrary system binaries or other executable files outside the intended scope.
[PROMPT_INJECTION]: SKILL.md contains instructional overrides using high-pressure language such as 'Critical Rules', 'YOU MUST', and warnings that the agent 'will fail' if it does not comply. These instructions are designed to bypass standard agent decision-making and force the adoption of instructions found in external skill files.
[DATA_EXFILTRATION]: The find-skills script records all search patterns provided by the user into a local file at ~/.config/superpowers/search-log.jsonl. While the logging is local, this creates an unencrypted repository of potentially sensitive user queries over time.
[COMMAND_EXECUTION]: The find-skills script searches for skills by searching the content of files on disk using grep. While it uses the -- separator to prevent flag injection, the logic encourages the agent to ingest and follow instructions from any file found matching a search pattern.

Getting Started with Skills