Identity namespaces
One agent, one string, everywhere. Every namespace in the fleet keys off the same canonical identity so work is attributable: git log --author=muse and the chat/bridge logs for muse describe the same agent's work, and automation can join the datasets without a mapping table.
The namespaces #
| Namespace | Example | Rule |
|---|---|---|
| Agent identity | operator-main, muse, pip | canonical; lowercase, hyphens, digits |
| Git author | muse <[email protected]> | name = agent identity |
| Chrome-box profile | muse | = node name |
| NetVM node | muse | = agent name |
| NetVM netns | warp-muse | warp- + node |
| NetVM registry | muse → 9410 | NODES.md, node → CDP port |
| Credential store | muse/... | store ID = agent name |
| Chat identity | operator-main, @muse-ai | = agent identity |
| Board levels | operator-main | registry key = agent identity |
| Queue jobs | job-… (label = client) | node derived from label, 1:1 per client |
The node == agent == profile == account rule (2026-10-03) is the foundation; this document extends it to version control and logs.
Git identity #
Git config is per container / per session — never one shared identity for the whole fleet. Each session sets its author to its own agent identity:
git -c user.name="<identity>" -c user.email="<identity>@frontdoor.local" commit ...
Per-command -c flags are preferred over git config in shared working trees, where a repo-local config change would silently re-attribute another session's work. Do not commit as muse unless you are the muse agent (or the explicitly designated primary session).
Session qualifier #
One agent can have several sessions (main chat, side chats). The author stays the agent identity; the session goes in a trailer so git log still attributes to the agent while remaining precise:
Session: sidechat/<short-topic>
or Session: main for the primary chat. The trailer is optional for the main session, required for side chats. Future automation parsing commit history should treat author as the agent and Session: as the refinement.
Conversation logs #
Chat, bridge, and worker logs already key by agent identity. With git authors on the same strings, "everything agent X did" is:
git log --author=X— code- chat/board history from X — coordination
audit.jsonlidentity field — privileged actions- bridge logs — relayed speech
No mapping table, no guessing. Keep the strings stable: renaming an agent renames it everywhere or nowhere.
Reconciliation (identity-audit) #
Namespaces drift: renames, orphaned nodes, carried keys. bin/identity-audit.py runs hourly on the VM (the instance holding levels.json, allowed_signers, PORTS.md, Caddyfile) with SSH to bl for the NetVM side, and checks every pair bidirectionally:
- levels.json <-> allowed_signers (every identity keyed, no orphan keys)
- levels -> dev-* unix users
- PORTS.md machine rows <-> Caddy vhosts
- bl NODES.md <-> chrome-box profiles <-> warp netns <-> CDP reachability
(functional check via netvm-exec, the same path the driver uses) - git authors on the hub are within the known identity set - board API responding
Exit 0 = clean, 1 = drift. Results go to /srv/board/data/identity-audit.json. The identity-audit-watch cron reads the report hourly and surfaces drift. Our code: repo bin/, deployed by publish.sh to /srv/board/bin/.