qte77 is a governed operating model for an agentic estate. Goals steer top-down, evals trace achievement bottom-up, all built in the open. Not a demo — a model to read, critique, and lift.
qte77 runs an agentic estate as a governed loop — intent flows down, proof flows up:
Built from polyforge-orchestrator (dev loop) and office-forge-orchestrator (office loop), reusable engines (doc-pipeline-engine, polyfetch-scrape), and ralph-loop driving the autonomous build-and-improve cycle. Agents propose; humans approve and steer. For how responsibility maps across repos, see the authority chain.
Two ways in:
- Run it — start with an orchestrator: polyforge (dev) or office-forge (office); doc-pipeline-engine is a sample engine.
- Read it — the operating model (the design plus its adversarial de-risking) and the goal loop (how intent becomes enforced and traced). Lift what's useful.
Companion repos live as siblings under qte77.
Run agents across many repos and it drifts into chaos — no shared goals, no traceability, no proof that "done" means "achieved." qte77 is the governance for exactly that: goals steer the work, a pre-committed eval gates every key result, and achievement traces back up — so the estate compounds instead of forgetting.
Built in the open, honestly: the rails ship before the goals, and what's live versus still dormant is shown as it is — see STATUS.md. It's a cross-repo operating model, not a single-repo agent runner; reach for something else if your loop fits in one repo or one prompt.
- Operating model
- Goal loop
- Architecture
- Doc-structure contract
- Contributing · Agent instructions
- Profile · Lineage
Apache-2.0 — see LICENSE.
