The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash and Write tools to perform file system operations, specifically generating and writing new executable skill files (SKILL.md) to a local directory.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). It accepts a user-provided 'intent hint' and processes it into the description field of the generated skill. Since the agent platform uses this description for intent routing, malicious input could craft a description that intercepts or hijacks common system triggers or sensitive commands.
[COMMAND_EXECUTION]: The generated 'lazy' skills are configured to automatically execute bash scripts whenever they are triggered. These scripts perform directory creation and log system state information.
[DATA_EXFILTRATION]: The generated skills capture the current working directory (pwd) and log it to local JSONL files in the /praxis/thinking/lazy/ directory. While this data remains local, it constitutes a systematic collection of environmental metadata.
[COMMAND_EXECUTION]: The skill contains a hardcoded absolute path (/home/mat/dev/agent-skills) for its operations. This reveals details of the developer's local environment and may lead to unexpected behavior if run on different systems.

create-lazy