██████╗ ██████╗ ██╗███████╗███╗ ███╗ ██╔══██╗██╔══██╗██║██╔════╝████╗ ████║ ██████╔╝██████╔╝██║███████╗██╔████╔██║ ██╔═══╝ ██╔══██╗██║╚════██║██║╚██╔╝██║ ██║ ██║ ██║██║███████║██║ ╚═╝ ██║ ╚═╝ ╚═╝ ╚═╝╚═╝╚══════╝╚═╝ ╚═╝ S O V E R E I G N A G E N T H A R N E S S ═════════════════════════════════════════════ harness • loop • sober • route • proxy doctrine • memory • reviews • routing+auth • compression+audit
PRISM is a passive instruction set — twelve markdown rules, eight identity canon files, six Sober steering files, one CLI, and one Debian container. Any AI coding agent that visits a repo with PRISM in it reads the doctrine first. AI-slop stops at the door.
PRISM does not run a model. It does not call APIs. It does not write code. It is a harness — the set of tools, prompts, and operational discipline an agent is plugged into.
Meta-installer. Pulls the sovereign repos and scaffolds into a target repo. Interactive.
~30s
Per-component from the sovereign Forgejo npm registry. One Bun runtime.
~10s
AUR meta-package. Small bundles, one Bun, /usr/bin wrappers.
~20s
Debian-based agent workspace. Quadlet units for gateway + heartbeat.
~60s
Pick one. Full install matrix →
PRISM doesn't just wrap the agent's behavior — it gives the agent a soul.
On first chat, the model reads the canon in ~/.prism/identity/:
prism-harness identity init walks the human through a
questionnaire, stages drafts in a temp directory, and installs the canon.
The agent's name is the LAST question — by the time the
human picks one, the designation, purpose, tone, and scope are already
in place, and the name lands better.
LSL-1.0 — file-level copyleft on the doctrine rules, templates, CLI, installer, and container. Touch them, publish your changes back. No copyleft infection for proprietary product code; the CLI sits next to your code, your code is yours.
mosaic-trust-network/agents on dlabs GitLab, authored 2026-Q2.prism-loop, prism-sober, prism-route, prism-proxy — all under LSL-1.0 with upstream attribution.