Standing

From its catalog listing
Earned trust for Hermes workers. Each profile has a level (L0 read, L1 draft, L2 act, L3 trusted) that decides which tool calls need the owner's approval, and a local record of checks that held up and checks that did not. A worker can always report on its own card (block, comment, request review, heartbeat), but it cannot mark the task done on its word: kanban_complete is blocked until the task has a passing standing_proof in the same session, and Standing runs that check itself (owner-scoped file, owner-allowlisted HTTP URL, or exact owner-allowlisted read-only command). Ten clean passes raise a level (capped at L1 automatically; L2 needs the owner's explicit opt-in); one failed check lowers it; a check that could not run is logged as unavailable and changes nothing. Seeds only create new actors, so restarts never undo a demotion. /standing shows the record, /standing log the last events, and /standing set (local owner only) changes a level. Optional same-machine shared ledger for a governor and its workers, and a short score line on the task card after a completion or demotion.
Good to know before installing
pre_tool_call returns Hermes' own approve directive for every tool call outside the profile's level (at the default L1 that includes every terminal/shell, send and unknown-tool call) and blocks kanban_complete until a Standing proof passes in the same session; a worker's block/comment/review/heartbeat calls on its own assigned card skip the Standing prompt; under cron and other unattended runs Hermes' approval policy (default deny) decides; it adds a guidance note to the model context through pre_llm_call; it never auto-approves or calls a model, and makes network or subprocess calls only for owner-scoped proofs (HTTP GET with no redirects, proxies or credentials; exact allowlisted argv, shell=False, minimal environment); it stores actors and hashed proof events in SQLite under its plugin data directory (or an owner-set db_path) and, when card_line is on, posts task-card comments through the public kanban_comment tool.
From its catalog listing.
How to install
- Install it. Hermes asks for any API keys it needs, then asks whether to turn it on.
hermes plugins install hermes-standing - Said no? Turn it on later:
hermes plugins enable hermes-standing - Restart Hermes so it loads the plugin.
Some plugins also need a sign-in or settings: see Full setup & settings. Memory plugins ask whether to become your memory provider instead. Steps follow the official plugins guide (opens in a new tab).
- Category
- Security & privacy · Automation & multi-agent
- Maintained by
- percyhendrux123-prog (Community)
- Needs
- No API keys declared
- Adds tools
standing_proof- Adds hooks
pre_tool_callpost_tool_callpre_llm_callon_session_finalizeon_session_reset- Works on
- linux, macos
- Needs Hermes
- 0.22.5 or newer
- Version
- 0.1.3
- In the catalog
- 2026-11-09, updated 2026-11-10
Known issues (from its catalog listing)
- A passing check proves only what it checks. Standing does not judge whether the whole deliverable is right; choose meaningful checks.
- Not a sandbox. Other internal hook errors fail open (logged, no directive); only the completion proof check fails closed.
- At L2, shell and unknown tools run without a Standing prompt so proven unattended workers can work; recognized spend and delete/move calls still ask below L3.
- Sandboxed workers cannot use host file proofs; use owner-allowlisted HTTP or command proofs instead.
- Command proofs are POSIX only (Linux, macOS).
Using the Hermes desktop app? See where plugins live

Full setup & settings ↗Source code ↗
More in security & privacy
All security & privacy plugins
New Hermes Agent updates every day on X. Follow @HermesWatcher