The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes session logs and persistent memories from the Hermes agent, which constitute untrusted external data. These sources could contain malicious instructions or prompt injections from previous interactions designed to subvert the distillation process.
Ingestion points: Ingests data from file paths defined in configuration, defaulting to ~/.hermes/memories/ and ~/.hermes/sessions/ (SKILL.md).
Boundary markers: The skill instructs the agent to summarize conclusions and patterns rather than quoting raw transcripts verbatim, which serves as a mitigation against direct instruction following from source material (SKILL.md, Step 3).
Capability inventory: Includes file system write access to the Obsidian vault and execution of the obsidian-wiki and qmd CLI tools (SKILL.md, Step 6 and QMD Refresh).
Sanitization: Explicit guidelines are provided to remove API keys, tokens, and credentials, and to redact private identifiers during the distillation process (SKILL.md, Step 3).
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage the Obsidian vault's state and search index.
Evidence: Calls obsidian-wiki memory sync with dynamically generated arguments and uses ${QMD_CLI:-qmd} for index updates and embedding generation (SKILL.md, Step 6 and QMD Refresh). These commands are integral to the skill's functionality but interact with the operating system environment based on processed summary data.

hermes-history-ingest