The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download code from a third-party repository not associated with trusted organizations.
URL: https://github.com/stephenschoettler/hermes-lcm
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of external code on the host system through the installation and update workflows.
Pattern: https://github.com/stephenschoettler/hermes-lcm → git clone into plugins directory followed by execution of ./scripts/install.sh or ./scripts/update.sh.
[COMMAND_EXECUTION]: The skill uses shell commands to manage its lifecycle and perform maintenance tasks.
Commands: Execution of ./scripts/install.sh, ./scripts/update.sh, and the lcm_doctor tool which supports destructive cleanup operations when LCM_DOCTOR_CLEAN_APPLY_ENABLED is set to true.
[INDIRECT_PROMPT_INJECTION]: The skill features a context recovery mechanism that is susceptible to indirect prompt injection from historical data.
Ingestion points: Tools such as lcm_grep, lcm_expand, and lcm_expand_query ingest data from a local SQLite database (lcm.db) containing prior conversation logs.
Boundary markers: None. The skill does not define delimiters or provide instructions to the agent to ignore potentially malicious instructions embedded in the retrieved historical context.
Capability inventory: The skill has the ability to write to the file system (externalized payloads), modify the SQLite database, and invoke an auxiliary LLM for synthesis.
Sanitization: There is no evidence of content sanitization or escaping of the historical data before it is re-interpolated into the agent's active context.

hermes-lcm-context-management