The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an orchestration workflow where an agent reads external files—including project source code, 'context packages' (context.md), and reference materials—and incorporates their content into instructions for sub-agents via the delegate_task tool. This ingestion of potentially untrusted data creates an attack surface where instructions embedded in those files could influence or override the sub-agents' behavior.
Ingestion points: External project files, context.md, PRINCIPLES.md, and local source checkouts (as specified in SKILL.md and the templates in references/).
Boundary markers: Templates utilize role headers (e.g., [omh-role:planner]), but the skill does not explicitly mandate the use of delimiters or 'ignore' warnings for the content of the ingested files.
Capability inventory: The orchestrator utilizes delegate_task for tool invocation, terminal access for git operations, and file system writes for artifact generation.
Sanitization: No explicit sanitization, escaping, or validation of external file content is mentioned before it is interpolated into sub-agent prompts.
[COMMAND_EXECUTION]: In SKILL.md under the 'P9' pitfall section, the skill provides a 'Pre-flight checklist' that instructs the agent to execute several shell commands (git remote -v, git fetch --all --tags, git rev-list, and git describe) in a terminal to verify the freshness of a local source tree before proceeding with design tasks.

omh-ralplan-driver