The Agent Skills Directory

[SAFE]: The skill establishes a read-only auditing workflow for repository surfaces (skills, prompts, plugins, config). It includes explicit safety rules to never print secret values, record only redacted information, and avoid mutating files or runtime states.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it is designed to ingest untrusted data from the local workspace.
Ingestion points: Reads repository files, skills, prompts, plugins, and configuration files (SKILL.md).
Boundary markers: Mentions a 'not-evidence boundary', though specific prompt delimiters for external content are not defined.
Capability inventory: Executes the omh CLI tool for runtime recording (SKILL.md).
Sanitization: Instructs the agent to redact secrets and maintain a read-only boundary. Given the skill's primary purpose is auditing, this ingestion is necessary and mitigated by the stated constraints.
[COMMAND_EXECUTION]: The skill uses a local CLI command omh runtime record to track the status of the audit. This is an operational command consistent with the tool's integration into the 'oh-my-hermes' (OMH) workspace environment.

omh-workspace-audit