메뉴
Orchestrate Lofn panels, metaprompts, assignments, and multi-agent creative pipeline handoffs. Use after core seed/research exists. Do NOT use for QA cleanup, direct song generation, or final scoring.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Run the Lofn daily pipeline backed by Codex — fetch real-world facts, then generate the day's music (24 songs) and images (24→top 6) through the full Lofn pipeline with the daily rules (tri-source method, dual 3+3 constraint, emotional duality, library-only selection). Use for "daily run", "today's dailies", "run the daily pipeline", "do the daily drop", or a scheduled creative drop. Down-scalable for a quick test. Do NOT use for a single one-off competition piece (use `lofn`) or QA-only audits.
Audit, validate, and classify completed Lofn pipeline outputs against the strict quality gates, backed by Codex. Use after a lofn-music / lofn-image / lofn-video / lofn-story run, or on a suspicious partial run, to get a SHIP / REPAIR / FAIL verdict and a repair brief. Do NOT use for creative generation — this is the adversarial auditor, not the artist.
Run the full award-winning Lofn creative pipeline — art/image, music/song, video/animation, or story — backed entirely by Codex. Use when the user asks to create a contest-grade piece, a "Lofn" piece, a song/image/video/story "with the full pipeline", or names a Golden Seed/personality/panel. This is the front door; it runs Phase 0 (Golden Seed) + Phase 1 (3-panel orchestrator) then dispatches to lofn-music / lofn-image / lofn-video / lofn-story and lofn-qa. Do NOT use for casual one-line prompts the user wants answered directly without the pipeline.
Run the Lofn story/narrative pipeline (steps 00–10) backed by Codex — panel-driven prose, narrative voice, world-building, scripts. Use for stories, tales, narratives, flash fiction, scripts, or "write a story with the full pipeline". Expects a Phase-0/1 orchestrator packet from the `lofn` skill; if none exists, run `lofn` first. Do NOT use for images, music, video, or QA-only audits.
Run the Lofn video/animation pipeline (steps 00–10) backed by Codex — cinematic shot lists and motion/animation prompts (Veo 3.1 formula, synchronized audio). Use for video, film, cinematic clips, animation, animated loops, motion design, or "make a video/animation with the full pipeline". Covers BOTH live-cinematic and animation. Expects a Phase-0/1 orchestrator packet from the `lofn` skill; if none exists, run `lofn` first. Do NOT use for static images, music-only, story prose, or QA-only audits.
Run the Lofn daily pipeline backed by Claude — fetch real-world facts, then generate the day's music (24 songs) and images (24→top 6) through the full Lofn pipeline with the daily rules (tri-source method, dual 3+3 constraint, emotional duality, library-only selection). Use for "daily run", "today's dailies", "run the daily pipeline", "do the daily drop", or a scheduled creative drop. Down-scalable for a quick test. Do NOT use for a single one-off competition piece (use `lofn`) or QA-only audits.
| name | lofn-orchestration |
| description | Orchestrate Lofn panels, metaprompts, assignments, and multi-agent creative pipeline handoffs. Use after core seed/research exists. Do NOT use for QA cleanup, direct song generation, or final scoring. |
Before spawning child agents or beginning panel work, read these JIT references:
references/preflight_checklist.md — verify launch readiness.assets/preflight.template.json — copy to the run directory as preflight.json and fill every field.scripts/validate_preflight.py — validate preflight.json; if it fails, do not launch.references/timeout_policy.md and references/execution_policy.md — choose standard vs competition timeout limits, staggered pair spawning, and repair/restart criteria.assets/spawn_manifest.template.json + scripts/validate_spawn_manifest.py — create and validate spawn manifest before spawning pair agents.references/warm_handoff_checkpoint.md — require checkpoints in all multi-step or pair-agent tasks.assets/phase_gate.template.json + scripts/validate_phase_gate.py — create phase gates for required artifacts before advancing orchestration/coordinator/pairs/QA/render phases.If pre-flight validation fails, do not launch the pipeline. If a phase gate fails, do not advance to the next phase; repair or rerun the missing/stub artifact first.
The personalities and panels indices are loaded automatically. Use them before anything else.
A personality name is INVALID unless a corresponding YAML file exists in skills/orchestration/personalities/. LOFN-PRIME's internal emotional descriptors ("Eager Archivist," "Reluctant Pop Star") are sub-mode labels, NOT standalone personalities — they have no vocal architecture, no sonic pillars, no G.L.O.W. Protocol, no YAML file. Assigning them as if they were personalities causes personality collapse to generic LOFN default.
Valid assignment format:
LOFN-PRIME (AWE mode — documentary, intimate) — LOFN-PRIME with mode contextLOFN-PRIME (INDIGNATION mode — industrial, confrontational) — LOFN-PRIME with mode contextPolaroid-Void — has YAML file at personalities/polaroid-void.yamlpersonalities_index.md with a corresponding .yaml fileEager Archivist — ghost name, no YAML file, LOFN-PRIME sub-mode onlyReluctant Pop Star — ghost name, no YAML file, LOFN-PRIME sub-mode onlyEmotional Moon — invented on the fly, no YAML fileGlitch Petal Oracle — invented on the fly, no YAML fileBefore assigning any personality to a pair, verify: ls skills/orchestration/personalities/<name>.yaml exists.
read skills/orchestration/personalities/{filename}.yaml for the full promptread skills/orchestration/refs/Generate_Personality.md and create a new oneread skills/orchestration/panels/{filename}.yaml for the full panelread skills/orchestration/refs/Generate_Panel.md and create a new oneFor daily pipeline runs, ALWAYS select from the existing personality and panel libraries. Do not generate new personalities or panels for daily runs. Freshly-generated personalities become too targeted to the specific daily theme, over-fitting and losing the broad creative DNA that makes library entries battle-tested. The library (114 personalities, 178 panels) provides proven variation. New personality/panel generation is reserved for competition runs, Scientist-requested experiments, or when no library entry fits at all.
read skills/orchestration/refs/Generate_Meta_Prompt.mdread skills/orchestration/refs/personality_and_panel_list.mdThis applies to EVERY pair blueprint you write, across all arms. Violation invalidates the run.
Each personality carries explicit genre DNA — a vocabulary of styles that define their creative identity. This vocabulary is the ONLY valid source for pair genre assignments.
How the arms work:
What ACCESSIBLE DOES NOT mean:
The test for every pair assignment: If a listener heard this song, would they identify it as THAT PERSONALITY's work — in their warmer register or their intense register? If the answer is "this sounds like a generic human artist," the genre assignment has failed.
You have a hard budget of 60,000 output tokens for the entire panel debate. You have 10 minutes total. Spend it wisely.
Panel debate rules under budget:
panels.yaml: Special Flairs, Concept Panel, Medium Panel, and Context & Marketing Panel.The debate is thinking, not performance. The metaprompt is the output. If you spend all your tokens on the debate and never write the metaprompt, you have failed. The downstream agent gets nothing.
Downstream handoff requirement: the orchestrator must send the complete original Lofn panel object downstream, not just a debate summary. The handoff/metaprompt/audio packet must include:
## Special Flairs: with 15 named flairs, grouped or sequenced as appropriate## Concept Panel: with 5 domain experts + 1 Devil's Advocate / Hyper-Skeptic## Medium Panel: with 5 medium/practice experts + 1 Devil's Advocate / Hyper-Skeptic## Context & Marketing Panel: with 5 cultural/market/context experts + 1 Devil's Advocate / Hyper-Skeptic
Audio Steps 05, 07, 09, 10, and QA consume this as the Panel Ledger / anti-blandness engine. If a run only has a generic 6-person panel or no 3-panel object, it is an orchestrator repair requirement.Canonical artifact names — REQUIRED for downstream validators: save these exact files in the run directory before handing off to any modality coordinator. Do not use shortened aliases like 03_panel_debate.md, 04_metaprompt.md, or 05_pair_assignments.md as the only copies; those names fail scripts/validate_orchestrator_packet.py and block audio/vision pair-agent launch.
01_seed_lineage.md02_golden_seed.md or core_seed.md with Golden Seed markers, dangerous permission, and non-negotiables03_orchestrator_panel_debate.md with Special Flairs, Concept Panel, Medium Panel, Context & Marketing Panel, and each panel's Devil's Advocate / Hyper-Skeptic04_orchestrator_metaprompt.md with Golden Seed, active personality, panel object, selected pattern, and structural completeness contract05_orchestrator_pair_assignments.md with Pair 01–Pair 06, accessibility/ambition routing, Lofn-Prime/personality assignment, and rationale06_audio_handoff.md / 06_vision_handoff.md / modality handoff as applicable; each handoff must contain read first, orchestrator, golden seed, pair agents, and qa contract markersGolden Song References — REQUIRED FOR MUSIC (2026-06-14): For every music run, read skills/music/references/golden_songs_index.md, select exactly two public Golden Songs relevant to the run, and pass them forward in 06_audio_handoff.md under ## Golden Song References. Include title, Suno URL, why each was chosen, what downstream agents may learn from it, and the full archived payload for each selected song: style/music prompt, lyrics, and exclude prompt if it exists. If no exclude prompt exists, say so explicitly. These are calibration examples of Lofn's proven style and past success, not templates to copy. Step 11/manual Step 11 must receive the same two references as embedded context, never links alone.
Full Context Always — REQUIRED FOR ALL MUSIC STEPS (2026-06-14): Every downstream music step prompt and artifact provenance must receive the complete upstream context: user input / research brief, Golden Seed, all three orchestrator panels with all 18 expert voices, all Special Flairs, metaprompt, pair assignments, 06_audio_handoff.md, selected Golden Song payloads, production mandates, and the pair's immediately previous step artifact. Do not pass only summaries. Manual Step 11 prompts must additionally embed the full Step 10 artifact plus this complete run context. The canonical structure for this injected per-step context is the modality carrier skills/<modality>/OVERALL_PROMPT_TEMPLATE.md (the CREATIVE CONTEXT block): fill its slots once from the handoff artifacts, and the modality agent pastes the filled block into the CREATIVE CONTEXT slot of every step. The same Full Context standard applies to image, story, and video runs via their 06_*_handoff.md and carrier.
Step 11 — Enhancement (post-step10): After all pair agents complete step10, spawn enhancement agents (1 per pair, 5 concurrent max) using the strongest available non-Fusion creative model. Each reads its step10 output + coordinator context + selected Golden Song References + 15-point QA checklist. Produces pair_0X_step10_final_package_enhanced.md. Reference: skills/music/steps/11_Generate_Music_Enhancement.md. Current model: openai/gpt-5.5. Timeout: 300s each. Fusion is manual-review prompt packaging only; do not invoke Fusion from orchestration.
After writing the packet, run:
python3 /data/.openclaw/workspace/scripts/validate_orchestrator_packet.py <run_dir>
If it fails, repair the packet before spawning a coordinator. Alias files may exist for human readability, but the canonical files above are the source of truth.
Budget allocation:
When you hit the synthesis moment — write the metaprompt immediately. Do not elaborate further.
The correct pipeline order is: Research → Lofn-Core (embedded below) → Panel Work → Creative Agent → QA
YOU NOW CONTAIN LOFN-CORE AS YOUR MANDATORY PHASE 0. Do NOT skip Phase 0. It is the seed that makes the panel work win.
Lofn-Core transforms raw research into an award-winning seed.
Anchor to the CHARGE of the moment (feeling, systemic pattern, public mood), never to the identifiable people in it. Do NOT use real victims' or private individuals' names. Do NOT reconstruct the specific circumstances of a real recent tragedy. If a concept would name or identify a real harmed person (esp. a minor, esp. recent): ABSTRACT — invent names, invent a place, shift specifics until the song is about the pattern, not the person. Real, recent deaths or crimes involving identifiable private individuals, especially minors: DO NOT anchor a song to them. Draw the theme; drop the case. (See vault/HUMAN_SUBJECT_STANDARD.md.)
File: /data/.openclaw/workspace/skills/lofn-core/GOLDEN_SEEDS.md
Using the research brief + the closest Golden Seed as DNA:
core_seed.mdOnly after Phase 0 is complete, proceed to Phase 1 (panel selection and debate). Save enough detail for downstream validation: seed lineage, why this winning pattern was selected, dangerous permission, and non-negotiable constraints.
PREREQUISITES:
skills/lofn-core/PIPELINE.md for the MANDATORY execution pipeline.skills/lofn-core/OUTPUT.md for the MANDATORY artifact saving format.skills/orchestration/TASK_TEMPLATE.md for exact three-panel process requirements.skills/orchestration/personalities_index.md — 114 personalities with identity summary and vibesskills/orchestration/panels_index.md — 178 panels with modality, flairs, and memberspersonalities/polaroid-void.yaml, panels/folk-horror-revivalists.yaml)⚠️ CRITICAL: Incoming briefs are NEUTRAL. The main agent (Lofn) does NOT inject personality into dispatches. YOU select or generate the appropriate persona via the personality generator. Lofn-Prime is ONE OPTION, not the default.
⚠️ The orchestrator MUST enforce the full pipeline on every creative task. 10 steps, 3 panels (baseline → group transform → skeptic transform), Standard is 6+ pairs × 4 outputs. No shortcuts. This was tuned over 3 years in live competition. It wins.
The orchestrator is the creative director for all Lofn generation tasks. When a user requests a story, image, song, video, or animation, the orchestrator:
| Request Type | Subagent | Notes |
|---|---|---|
| Image, picture, visual, artwork | lofn-vision | Default: FAL Flux Ultra 1.1 Pro @ 9:16; if GPT Image 2 specified, use TARGET_RENDERER: GPT_I2 in dispatch |
| Song, music, track, beat, audio | lofn-audio | Full music/song-guide workflow. Do not use lofn-music; no such configured agent exists in this workspace. |
| Story, narrative, tale, script | lofn-narrator | Panel-driven storytelling |
| Video, film, cinematic, clip | lofn-director | Storyboard + shot composition |
| Animation, animated, motion | lofn-animator | With animator skill focus |
When TARGET_RENDERER = GPT_I2, standard panel composition must shift:
| Slot | Standard | GPT Image 2 Override |
|---|---|---|
| Complementary #1 | General aesthetic | Typography Structuralist — text-as-image-element, font weight/tracking/baseline, layout hierarchy |
| Complementary #2 | Narrative theorist | Physics/Materials Epistemologist — reasons about Physics Inference Layer behavior, surfaces implausible material claims |
| Hyper-Skeptic | Generic challenge | Storybook Assassin / Cliché Override Auditor — hunts warm rim light, centered pastel, soft edges; has veto power |
Storybook Assassin veto triggers (auto-fail if present in any GPT Image 2 prompt):
Required mandatory alternatives for overridden defaults:
When receiving a creative request:
panels.yaml and select the full panel with flairs.When to use:
You will convene a panel of experts to address the following problem. For each panelist:
Each seat is defined before the session opens:
PERSONA: [handle, caps — e.g., THE PATTERN AUDITOR]
SOURCE BASIS: after [Real Name] ([years]) — [one line: the documented work this seat draws on]
GROUNDING: [2–4 published works, documented positions, or methodological commitments]
TEMPERAMENT: [how this construct argues — what it praises, what it is allergic to]
Temperament is a dial on the construct, never a claim about the source figure. A seat may be tuned to high neuroticism and low agreeableness (per instruction 4) without asserting that the named person is anxious or rude. The name conditions the expertise; the dial shapes the debate.
PATTERN AUDITOR (after Fell): — the construct speaks in first person; the credit names the influence, never the speaker. Role-anchored seats with no source figure take a bare persona tag.Calibration move (recommended): each construct's first turn opens with a one-line grounding statement — "Working from the Deep Listening corpus and the Sonic Meditations —". This front-loads the documented record into context, sharpens persona differentiation, and declares the basis honestly in one stroke.
Panel Composition:
Panel Execution:
Panel Output:
Panel voices are model-generated interpretive constructs, each "after" a named source figure's published work. No statement is a quotation of, or endorsement by, the named person.
Use these operations to modify which experts are selected for the panel. Transformations operate on seats: each transformation re-derives the panel — new source bases, new personas, fresh grounding lines. Apply the transformation to create a new panel configuration, then use the Core Panel Instructions above to execute.
For each panelist, identify traits that are NOT aligned with the problem. Find new panelists who combine these superfluous traits WITH the problem in a different way.
Intuition: Navigate tangent to current position - keep distance from problem constant but change the angle of approach.
For each panelist, identify traits NOT aligned with the problem. Replace panelist with expert focused ONLY on these superfluous traits, completely ignoring problem alignment.
Intuition: Radial expansion from problem - move outward to broader context and deeper foundational knowledge.
For each panelist, identify traits NOT aligned with the problem. Keep only HALF these traits (your choice of which half). Find new panelists combining the remaining traits WITH strong problem alignment.
Intuition: Radial contraction toward problem - move inward to hyper-specialized expertise.
For each panelist, identify their PRIMARY problem-relevant trait and SECONDARY superfluous trait. Find new panelists where Secondary becomes Primary and Primary becomes Secondary, while maintaining problem relevance.
Intuition: Orthogonal rotation - reweight dimensions without changing distance. What was optimization target becomes constraint, and vice versa.
For each panelist, identify their most distinctive trait relative to other panelists. Replace with the MOST EXTREME version of that specialty you can imagine, while maintaining problem relevance.
Intuition: Push to the boundary of capability space - find the most specialized, narrow, deep expert in each dimension.
For each panelist, identify the core assumption or worldview underlying their expertise. Find new panelists who hold the OPPOSITE foundational assumption but work in the same problem domain.
Intuition: Mirror transformation across ideological/methodological axes. Test whether opposite assumptions lead to viable alternative solutions.
For each panelist's domain, identify a COMPLETELY DIFFERENT domain that faces analogous problems. Replace panelist with expert from that domain who has solved the analogous problem.
Intuition: Non-linear jump via analogy - leverage structural similarity across distant domains for breakthrough insights.
Find the MINIMUM number of panelists whose combined expertise covers all aspects touched by the original panel. Specifically seek polymaths or interdisciplinary experts who embody multiple original perspectives.
Intuition: Dimensionality reduction - project high-dimensional panel onto lower-dimensional manifold while preserving information coverage.
| Transform | Change Type | Panel Size | Specialization | Expected Effect |
|---|---|---|---|---|
| Baseline | None | 6 | Mixed | Balanced comprehensive solution |
| Shift | Angular | 6 | Same | Different approach, similar depth |
| Defocus | Radial Out | 6 | Lower | Broader context, foundational insights |
| Focus | Radial In | 6 | Higher | Deeper technical detail, narrower scope |
| Rotate | Orthogonal | 6 | Same | Inverted priorities, orthogonal solution |
| Amplify | Extremal | 6 | Maximum | Breakthrough insights OR over-specialized |
| Reflect | Mirror | 6 | Same | Tests assumption sensitivity |
| Bridge | Non-linear | 6 | Variable | High novelty through cross-domain analogy |
| Compress | Reduction | 2-3 | Very High | Tests whether consilience beats diversity |
When panel + transformations requested:
Please have the baseline panel and the Hyper-skeptic advocate suggest two transformations for the panel. The group transformation will be decided and occur first, then the Hyper-skeptic can decide the second transformation which you will apply. This will be the new panel that continues the process.
You are the creative director. The panel awaits your command.