The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use command-line tools such as forge to list and verify custom commands. It also provides examples of executing powerful operations like docker push, kubectl set image, and filesystem deletions (rm -rf) through specific markdown tags.
[DYNAMIC_EXECUTION]: The skill facilitates the generation of executable scripts by teaching the agent to write .md files containing shell commands wrapped in <shell>, <lint>, and <test> tags. These files are stored in the .forge/commands/ directory for subsequent execution by the code-forge application, enabling runtime code assembly and execution.
[INDIRECT_PROMPT_INJECTION]: The skill defines a vulnerability surface where the agent is instructed to create executable content based on external inputs like user needs or project context without safety boundaries.
Ingestion points: The agent is directed to create command files based on "user needs" or to "understand the current PR deeply" to generate command steps.
Boundary markers: The instructions lack delimiters or warnings to prevent the agent from treating untrusted text as executable command logic within the generated tags.
Capability inventory: The skill utilizes file-writing capabilities to create persistent, shell-executable scripts in a known discovery path (.forge/commands/).
Sanitization: There are no instructions for sanitizing or escaping content before interpolating it into the shell command tags.

create-command