The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user descriptions and repository content (such as .PLAN.md or git logs) to generate agent instructions. Ingestion points include user input for command names and descriptions, as well as local repository files and git metadata. The skill provides a structured markdown template for generated commands, providing basic separation of instructions but lacks explicit input sanitization. It uses the Write tool to create executable command files and the Bash tool for repository interaction.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill and its reference patterns utilize the Bash tool for routine development operations, including directory management (mkdir -p), git repository status checks, and running testing/linting tools (make, pytest, ruff). These actions are consistent with the skill's purpose of automating developer workflows.
[DATA_EXPOSURE_AND_EXFILTRATION]: Example commands within the skill documentation reference tools like the Graphite CLI (gt) for submitting pull request stacks. This activity targets well-known development services (Graphite/GitHub) as part of the intended automation functionality and does not exhibit malicious exfiltration patterns.

command-creator