What you are looking at. This page shows the current work schedule of LIRIL, acting as a multi-role dev team. The user (CEO Daniel Perry) directed on 2026-04-18 that LIRIL should be autonomously writing and maintaining this schedule as if it were a full dev team. This page is that directive, made visible.
How each task is handled. LIRIL picks the highest-priority pending task, then routes it through six roles in sequence: a researcher pulls a primary source, an architect decides where it belongs, a designer picks the UX form, an engineer writes the literal content, an editor copy-checks, and a gatekeeper (the on-GPU hallucination gate) returns VERDICT: PASS / WATCH / FAIL. Only PASS reaches the commit stage. The installed commit-msg hook additionally rejects any jargon subject line.
Why this exists. Prior autonomous systems on this machine produced 1,357 commits in seven days labeled "[PHASE 14]", "ABCXYZ Millennial Falcon metrics", "Atmosphere optimizer matrices applied" — zero value. This dev team replaces that with a structured pipeline whose every step is auditable. See docs/AUTOMATION_DIAGNOSTIC_2026-04-18.md for the full before-state.
Team composition
In progress
Loading…
Backlog
Loading…
Completed today
Loading…
mercury.infer → llama-server (RTX 5070 Ti, ~2–8s per role). Classification uses tenet5.liril.classify on the NPU (~3–80ms, 99% confidence). All routed via host NATS on 127.0.0.1:4223 — the port fix that unblocked the documented daemon stack on 2026-04-18.Commit policy. Every commit from this dev team must pass the
.git/hooks/commit-msg jargon filter. The hook rejects subject lines containing [PHASE N], [LIRIL PHASE N], ABCXYZ, Millennial Falcon, Empirical Magic Handoff, Atmosphere optimizer, EMH tracker badge, Genocide-Evidence EMH, ActionGuard-verified, N-vs-NP convergence, and the mass-inject pattern. Hook installer: docs/install-commit-msg-hook.sh.Canonical data. data/liril_work_schedule.json is the on-repo source of truth for this board. The daemon at
E:\S.L.A.T.E\tenet5\tools\liril_dev_team.py reads and writes it atomically. Per-task transcripts land in data/liril_dev_team_log/ for reviewer audit.