The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a set of terminal commands (hermes sessions list, rename, archive, delete, prune, export) to manage the session library. These operations are restricted to the local session database and are governed by strict procedural rules requiring user consent.
[DATA_EXPOSURE]: The skill accesses the conversation message store via the session_search tool to summarize decisions and goals for the user. This is an intended primary function of the skill and does not involve exfiltration to external systems.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from past conversation logs, creating a potential surface for indirect injection if those logs contain malicious instructions.
Ingestion points: Data enters the context via the session_search tool results from the message store.
Boundary markers: The instructions specify a structured reporting format (link, goal, outcome) to help delimit session data from agent instructions.
Capability inventory: The agent can perform file-system and database mutations via the hermes sessions CLI (rename, archive, prune).
Sanitization: The skill mitigates risks by enforcing a mandatory 'Plan before acting' phase and requiring user confirmation for all mutations, including the use of --dry-run for destructive actions.

session-librarian