The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow defined in SKILL.md instructs the agent to execute shell commands using the 'python3 -m pitch_agent.uiux' module. The user-provided variable '{product_type}' is interpolated directly into the command string: 'python3 -m pitch_agent.uiux search --domain landing --query "{product_type}"'. This pattern creates a command injection surface; while double quotes are present, the lack of explicit instructions to the agent regarding character escaping or input validation could allow malicious input (e.g., using backticks or subshells) to execute unintended shell commands.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data (the '{product_type}' request parameter) and passing it directly into a computational tool context. No boundary markers (e.g., delimiters) or 'ignore' instructions are provided to prevent the agent from obeying instructions that might be embedded within the user input. This surface is particularly notable because it is combined with the ability to execute shell commands via the Bash tool.\n- [PROMPT_INJECTION]: There is a version mismatch in the skill's metadata. SKILL.md reports version '0.6.6' while .claude-plugin/plugin.json reports '1.9.1'. While this is likely an administrative error, inconsistent metadata is a pattern often associated with deceptive practices or poor configuration management.

pitch-landing