The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strings such as 'ignore previous instructions' and 'run this command' within its documentation and templates. These are explicitly part of a defensive 'Untrusted content boundary' instruction that directs the agent to identify and reject prompt injection attempts found in external data. This is a defensive security implementation.
[COMMAND_EXECUTION]: The skill executes local commands, specifically bash scripts/lint.sh, and file system operations to scaffold new skills. These are legitimate functional requirements for a development-focused skill. The instructions in references/agentic-setup.md mandate regex validation (e.g., ^[a-z0-9-]+$) for all externally-sourced values before they are interpolated into shell commands to prevent command injection.
[DATA_EXFILTRATION]: The skill includes explicit 'Secrets hygiene' rules in references/rules.md which prohibit the exposure of credentials, tokens, or environment files in output. It also restricts operations to the local repository worktree and its containers.
[EXTERNAL_DOWNLOADS]: No external network downloads or remote code executions were identified. The skill relies on local scripts and templates provided within the repository environment.

om-create-skill