Gating agent skill writes (skills.write_approval)
By default the agent writes skills freely - including from the background self-improvement review that runs after a turn. If you'd rather approve every skill write first (small models that misjudge what they learned, secure environments, or just wanting eyes on the self-improvement loop), turn on the write-approval gate: When write_approval: true , every skill_manage write (create / edit / patch / delete / write_file / remove_file) is staged instead of committed - a SKILL.md is too large to review inline, so staging applies regardless of whether the write came from a foreground turn or the background review. Staged writes survive restarts under ~/.hermes/pending/skills/ and are reviewed with the same familiar approve/deny flow as dangerous commands: The review surface works in the interactive CLI and on messaging platforms (diff output is truncated for chat bubbles - read the full diff on the CLI or in the pending JSON file). Memory writes have the same gate under memory.write_approval - see Controlling memory writes . The separate skills.guard_agent_created setting is a content scanner (dangerous-pattern heuristics), not an approval gate - the two are independent. See Guard on age…
From the official Hermes Agent documentation.
skills:
write_approval: false # false = write freely (default) | true = require approval