// documentation
llm-armory documentation
llm-armory holds named loadouts (executor lanes) you pull for different kinds of work. Fable (Claude Code) stays the advisor; you arm a session by choosing a loadout — primarily grok-high (Grok 4.5 at effort high).
Site: armory.muslewski.com · npm: llm-armory
Start here
| Path | For |
|---|---|
| Getting started | Install → armory --list → doctor → first armed launch |
| Works with | Fleet siblings (sage, herald, oracle, atlas, ferry) |
| Recipe: agent status | Launch stamps for co-installed judges / curtains |
Doctrine (short)
- Standalone-first — loadouts launch the same way with or without siblings.
- Explicit arm — the advisor chooses a loadout; pure Grok sessions use native
spawn_subagent. - Stamp once, then exec — Agent Status records are written immediately before
exec; soft-fail never blocks launch. - Loose coupling — readers (herald, sage, oracle) own their own truth after launch.
Where other knowledge lives
| Kind | Location |
|---|---|
| Public product docs | docs/ (this tree) |
| Architecture mind (Atlas) | llm-armory-mind/ — zones, decisions; specs/plans pipeline |
| Agent install / status runbook | AGENTS.md |
| Agent Status writer side | INTEROP.md |
| Human README | README.md |
| Changelog | CHANGELOG.md |
Agent design notes under docs/superpowers/ are historical; new specs/plans go to the mind vault (llm-armory-mind/specs/, llm-armory-mind/plans/).