The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided inputs to generate SKILL.md files, Python scripts, and test fixtures. It lacks explicit boundary markers or sanitization logic to prevent malicious instructions embedded in the 'working solution' from influencing the generated skill's behavior or tests.
Ingestion points: Processes 'working prototypes' or 'ad-hoc solutions' (Step 1, Step 2).
Boundary markers: None explicitly defined in the instructions for script generation.
Capability inventory: The skill uses Bash, Write, and Edit to create files and execute commands (Step 2, Step 4, Step 5).
Sanitization: No mention of escaping or validating input before interpolation into scripts or YAML files.
[COMMAND_EXECUTION]: The skill uses Python and shell commands to validate configurations and run tests. While these target local skill paths, they are executed based on user-provided skill names.
[DYNAMIC_EXECUTION]: Step 2 involves 'Deterministic extraction' which generates new Python or Shell scripts based on observed logic and then executes them to verify functionality. This process of automated script generation and execution presents a risk if the source logic is influenced by an attacker.

skillify