The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data to synthesize recommendations and draft durable records.\n * Ingestion points: The skill explicitly requires inputs such as goal and known evidence from the user (found in SKILL.md).\n * Boundary markers: Robust mitigation is present in SKILL.md and references/decision-records.md, which state that nothing is written to the file system until the user explicitly approves the draft.\n * Capability inventory: The skill has the capability to write markdown files to the docs/adr/ directory.\n * Sanitization: There is no explicit mention of sanitizing or escaping the user-provided evidence before it is interpolated into the strategy brief or ADR draft.\n- [COMMAND_EXECUTION]: The skill utilizes a project-specific CLI tool for internal status recording.\n * Evidence: The skill executes omh runtime record --skill strategy-brief --harness strategy-synthesis --status started as part of its runtime evidence collection in SKILL.md. This appears to be a legitimate telemetry or logging mechanism for the oh-my-hermes framework.

omh-decide