simnova

bothy-board

by simnova

Coordinate coding agents on BothyBoard — fail-closed dequeue over MCP. Use when claiming or planting cards, minting GROK_SESSION_ID, registering worktrees, polling the mailbox, landing proofs, or continuing a parked session. Triggers: BothyBoard, bothy-board_tasks_next, Planted, CAS claim, cacheToken, resume_from, worktree registry, proofs.set, treatments.fail.

finish-before-done

by simnova

"Turn-end completeness gate. Before reporting an activity done, re-read your own draft for sentences that DESCRIBE work rather than doing it - 'one gap worth naming', 'still needs', 'not yet', 'worth a look', 'I could not confirm', 'carried forward' - and for each ask whether you already have what you need. If yes, do it before sending; naming it instead is handing the operator work you could have finished. Carries the boundary that separates a real blocker from stopping short, and the create-only-field trap that turned one such omission into a rebuild."

board-conduit

by simnova

"Board-as-conduit protocol: GitHub project 7 cards are the owner-authored instruction SSOT consumed unaltered via briefFromIssue - card anatomy, lane definitions (A learner/XR, B API/admin), parent/child decomposition, concurrency N-gate criteria, worker staging limits (<300 turns), and the delegator/owner division of every dispatch step. Load BEFORE creating any project-7 issue/card, dispatching any worker, or changing concurrency."

openclinxr-openclaw

by simnova

Use when working in /Volumes/files/src/openclinxr on OpenClaw-style / OpenClaw-inspired autonomy, repo-agent consultation, case-definition-driven XR factory slices, coordination files, heartbeat replacement, or cross-harness Codex/Grok alignment.

OpenClinXR

by simnova

supervisor-loop

by simnova

One iteration of the four-duty supervisor loop — review agentic logs for non-self-correcting issues, keep the prioritized product-forward ready set from starving a dequeue, re-verify work reported done, and issue corrections. Load at the start of every supervisor iteration, before reading any log by hand.

orchestrator-dispatch-loop

by simnova

The CEO-side board loop — plant a RED, probe it, commit, dispatch, verify, integrate, close. Exact signatures and the traps that cost real time. Read before dispatching or harvesting a slice; the worker-side contract is delegated-worker-contract.

makeclothes-garments

by simnova

"Fit a MakeHuman .mhclo garment onto a body through the MakeClothes station. Load BEFORE acquiring a garment, running a fit, or refusing an asset on its licence header - carries the installed toolchain, the measured library-vs-worn gap, the hm08-versus-MPFB trap, and the licence precedence that reversed two wrong refusals."

consult-session-hygiene

by simnova

"Running long-lived superagent/peer consult threads without losing them. Token-per-call retirement budget, the no_visible_content fire alarm, why resume pricing is regressive, and the restore-stall signature. Load before resuming any consult thread or when a consult goes quiet."

OpenClinXR

by simnova

autonomy-boundaries

by simnova

"Gate for EVERY question, permission request, or 'may I / should I / which option / am I blocked' impulse, and for install / push / publish / merge impulses - the pre-approved table that means ACT, DO NOT ASK (install tooling, push main, publish, merge verified work - asking again IS the failure), then the escalate-last chain repo -> standing ox advisory thread -> one bundled operator message. Carries the live thread id, --resume invocation, recall verification, and thread-loss recovery. Load at the URGE to compose any operator question, not at the send."

grok-worker-monitoring

by simnova

How to watch a dispatched grok worker, converse with it across checkpoints, and route work to the right model. Three observation planes and what each can answer - the unified log is the only one that records CAUSES. Model capability table with verified capabilities and measured failure modes. Read when a worker looks stalled, before choosing a model, or before resuming a thread.

agent-session-continuity

by simnova

Pointer to the canonical session/resume rules. A wrong session id CONFABULATES rather than erroring; killed dispatches write no ledger entry; resumes need three env vars. Canonical text lives in PROTO_VERIFY_DELEGATION — this exists only because that file is 4,000 lines deep and stopped binding.

gh-body-file

by simnova

Never pass issue/PR/comment text to gh via --body "...". Backticks in the string are shell command substitution and silently mangle the published body. Always --body-file from a quoted heredoc, then verify.

trellis-vr-equipment-optimize

by simnova

Use when optimizing TRELLIS Metal equipment GLBs for WebXR/Quest budgets, multi-view Grok packs, factory:trellis:bake, or post-opt ladders. Codifies measured findings (photoreal vs hard-surface packs, meshopt plateaus, multi-view conditioning) and the proven iteration loop for VR prop budgets.

openclinxr-slice

by simnova

Rubric-scored evaluator for completed OpenClinXR work slices. Scores filesystem artifacts (slice record, evidence paths, package tests, guardrail deltas), not agent claims. Use when finishing a slice, verifying slice quality, or running skill:slice:check / test:skill:slice.

worker-scoped-session

by simnova

Run delegated headless/--yolo workers with OPENCLINXR_WORKER=1 so SessionStart docs hygiene and CEO coord hooks NO-OP. Use when spawning worktree workers, deepseek execute sessions, or any non-CEO agent that must not mutate registries/PROJECT_STATUS.

delegated-worker-contract

by simnova

Use for ANY delegated implementation/coding task in this repo — whether you are a headless grok/deepseek --yolo worker, a spawned subagent, or an orchestrator handing work off. Triggers on "implement", "add", "fix", "build", "refactor", "author", or any scoped code task run under --cwd/worktree isolation. Defines the non-negotiable execution contract. Proven (agentic-eval: skills enforce adherence better than --rules text).

large-task-orchestration

by simnova

Decompose large multi-file tasks into N disjoint file-scoped workstreams with worktree isolation, distinct portless ports, and per-job temp paths. Use when fan-out, parallel workers, large refactors, multi-package builds, Blender/mesh batches, or when the agent would otherwise solo a large task on the frontier model.

per-job-temp

by simnova

Unique per-pid/per-mesh temp paths for parallel asset/Blender/Python jobs. Use when writing under /tmp, sharing caches, multi-mesh skin/albedo, or any parallel file-producing pipeline to prevent output corruption.

regen-score-report

by simnova

Headless-friendly regenerate → score → report loop for OpenClinXR factory/evidence work. Use for model-vetting, orchestrate_character, garment/skin regen, or any bounded regen-then-measure cycle under OPENCLINXR_WORKER=1.

anny-asset-pipeline

by simnova

Use when working on OpenClinXR Anny or Anny-compatible humanoid generation, source import, Blender rigging/cleanup, provenance, preflight reports, or promotion gates for case-defined humanoid actors. Keeps Anny output as a candidate source until license, provenance, rig, actor-role mapping, WebXR evidence, and false-readiness gates are satisfied.

provider-boundary

by simnova

Use when OpenClinXR work involves local-only or approval-gated asset/model providers such as Anny, Hunyuan3D, Meshy, Tripo, ComfyUI/StableGen, VLM reviewers, voice/model providers, credentials, paid APIs, network calls, or provider execution gates.

antd

by simnova

Use when the user's task involves Ant Design (antd) — writing antd components, debugging antd issues, querying antd APIs/props/tokens/demos, migrating between antd versions, or analyzing antd usage in a project. Triggers on antd-related code, imports from 'antd', or explicit antd questions.

turborepo

by simnova

Turborepo monorepo build system guidance. Triggers on: turbo.json, task pipelines, dependsOn, caching, remote cache, the "turbo" CLI, --filter, --affected, CI optimization, environment variables, internal packages, monorepo structure/best practices, and boundaries. Use when user: configures tasks/workflows/pipelines, creates packages, sets up monorepo, shares code between apps, runs changed/affected packages, debugs cache, or has apps/packages directories.