The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is a wrapper for the 'ntm' command-line utility, designed to dispatch caller-provided shell commands to managed persistent panes. It executes these commands as a core functional requirement to facilitate agent tasks in terminal environments.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface through the 'pane-command-request' input, which contains external commands to be executed. Although the skill instructions emphasize factual reporting and clear boundaries, it provides a conduit for instructions from the agent's context to reach the system shell without explicit validation or sanitization within the skill logic itself.
Ingestion points: Untrusted data enters the context via the pane-command-request consumer defined in the frontmatter of SKILL.md.
Boundary markers: The instructions specify a 'One-shot dispatch' protocol and an 'Output contract' to maintain factual reporting, though they do not include explicit 'ignore embedded instructions' warnings for the executed data.
Capability inventory: The skill uses ntm to dispatch commands, which results in subprocess execution on the host environment.
Sanitization: There is no evidence of command sanitization or escaping within the skill body; it relies on the caller and the ntm tool to handle command safety.

ntm