The Agent Skills Directory

[NO_CODE]: The skill consists entirely of Markdown instructions and YAML/TSV configuration files. There are no executable scripts (Python, Node.js, Shell, etc.) provided within the skill package.
[SAFE]: The instructions emphasize 'least privilege' and 'bounded execution.' It explicitly directs the agent to ask for authority before performing external writes, destructive actions, or live-system changes. It also mandates that sensitive values should not be exposed during work evidence collection.
[COMMAND_EXECUTION]: While the skill mentions running local checks (compiling, running tests, CLI harnesses), these are standard development tasks performed in the user's local context. The instructions include safety guidelines to avoid sending credentials or destructive commands into test harnesses.
[DATA_EXFILTRATION]: The skill instructions specifically require that no sensitive values be exposed during the inspection of target states and that private artifacts (screenshots, transcripts) require explicit authority to keep.
[INDIRECT_PROMPT_INJECTION]: The skill processes 'work requests' or 'plans' as input. However, the instructions provide clear boundaries, requiring the agent to confirm targets and scope before action and to return to a 'planning' or 'understanding' phase if instructions are unclear or conflict with context rules.

kata-work