The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to access a sensitive file path to retrieve configuration data.
Evidence: The instructions direct the agent to check the ${HERMES_HOME:-~/.hermes}/.env file for the OBSIDIAN_VAULT_PATH environment variable (SKILL.md). Accessing .env files can expose sensitive secrets or other environment credentials to the model context.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from Obsidian notes which could contain malicious instructions designed to manipulate the agent.
Ingestion points: Note content is ingested via read_file and content-based search_files operations (SKILL.md).
Boundary markers: The skill instructions lack delimiters or explicit warnings to isolate untrusted note content from system-level instructions.
Capability inventory: The skill possesses file writing (write_file), file modification (patch), and shell execution (terminal) capabilities.
Sanitization: No specific sanitization or filtering of note content is mentioned before it is processed or used in operations.
[COMMAND_EXECUTION]: The skill explicitly permits the use of shell commands for path resolution and file operations.
Evidence: The instructions state that the terminal tool is acceptable for resolving environment variables or performing file appends when standard file tools are not applicable (SKILL.md).

obsidian