cellcog-multimodal
Orchestrate any-to-any multimodal sub-agent delegation via CellCog to generate 3D models, video, audio, executive documents, or citation-backed deep research. Do not use for basic single-file text editing.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- cellcog-multimodal
- description
- Orchestrate any-to-any multimodal sub-agent delegation via CellCog to generate 3D models, video, audio, executive documents, or citation-backed deep research. Do not use for basic single-file text editing.
02 资源文件 4
- 参考文档CARD.md3.4 KB
- 参考文档config.default.yaml294 B
- 参考文档README.md1004 B
- 参考文档REFERENCE.md4.4 KB
03 SKILL.md 正文
CellCog Multimodal Sub-Agent Delegation
cellcog-multimodal is the authoritative execution blueprint for delegating non-code and cross-modality tasks (3D modeling, video generation, audio/music synthesis, executive PDF/slide creation, and multi-source deep research) to CellCog via plugin.cellcog.
Every delegation cycle follows a five-stage progression:
[1. Task Assessment] → [2. Prompt & Tag Protocol] → [3. The Visual Brief] → [4. Mandatory Checkpoint] → [5. Execution & Verification]
See CARD.md for the skill summary card, quick-reference template, and completion criteria checklist.
Consult REFERENCE.md for copy-paste golden recipes across the top 5 modalities.
Consult /crafting-skills for authoring standards and /epistemic-isnad-audit for provenance tracking.
1. Task Assessment & Modality Selection
Evaluate the user's generative request to select the target operating point (chat_mode, chat_tier):
-
Modality & Mode Mapping:
chat_mode="agent": Most tasks — 3D model generation (.GLB), video rendering, audio/podcast synthesis, multi-tab Excel financial models, and code.chat_mode="creative": Design-taste work — interactive HTML dashboards, UI wireframes, brand identity kits, presentation slide decks, creative writing.chat_mode="team": Deep multi-source research ONLY — cross-validation, competitive matrices, and citation-backed synthesis.
-
Quality Tier Selection:
chat_tier="flash": Default fast/economical tier for light asset generation, quick scripts, or preliminary drafts.chat_tier="core": Balanced tier for standard decks, spreadsheets, and illustrations.chat_tier="max": Maximum depth tier for production video rendering, 3D character rigging, complex data analysis, or legal drafting.
Completion criterion: Target modality identified,
chat_modeandchat_tierselected with rationale.
2. Prompt Construction & Tag Protocol
Construct the delegation prompt using the dual tag protocol:
-
Input Grounding (
<SHOW_FILE>):- Wrap reference files (PDFs, images, code files, CSV data, audio) in
<SHOW_FILE>/absolute/path</SHOW_FILE>tags. - Security Invariant: Never wrap credentials,
.env,.git, or private keys in<SHOW_FILE>tags.
- Wrap reference files (PDFs, images, code files, CSV data, audio) in
-
Deterministic Output Path (
<GENERATE_FILE>):- Wrap desired output file paths in
<GENERATE_FILE>/workspace/output/artifact.ext</GENERATE_FILE>tags. - This guarantees artifacts download directly to deterministic locations rather than ephemeral temp folders.
- Wrap desired output file paths in
-
Explicit Artifact Declarations:
- State exact file formats and deliverable types explicitly (e.g. "Create an interactive HTML dashboard, a 60s summary video, and a formatted PDF").
Completion criterion: Prompt assembled with verified absolute
<SHOW_FILE>inputs and<GENERATE_FILE>output destinations.
3. The Visual Brief
For non-trivial multimodal pipelines, generate an interactive visual summary:
- Location: Write to
%TEMP%\cellcog-delegation-<timestamp>.html(Windows) or/tmp/cellcog-delegation-<timestamp>.html(Unix). - Topology Diagram: Render a Mermaid graph displaying:
- Input files (
<SHOW_FILE>) $\rightarrow$ CellCog Cloud Sub-Agent (chat_mode+chat_tier) $\rightarrow$ Output Artifacts (<GENERATE_FILE>).
- Input files (
- Surface: Deliver the absolute, clickable HTML file path to the user.
Completion criterion: Visual brief written to
%TEMP%and presented to user.
4. Mandatory Checkpoint
Before dispatching expensive cloud sub-agent tasks, present the delegation plan:
- Create or update
implementation_plan.mdartifact detailing:- Target modality and deliverables list.
- Selected
chat_modeandchat_tier. - List of attached reference files and target output paths.
- Set
RequestFeedback: truein artifact metadata. - STOP and wait for explicit user review and approval before executing the API call.
Completion criterion: Explicit user approval received for sub-agent execution.
5. Execution & Verification
Dispatch the task through the typed plugin.cellcog tools:
- Invoke Plugin Entrypoint:
- For general tasks: Call
cellcog_run(prompt=..., chat_mode=..., chat_tier=...). - For deep research: Call
cellcog_research(topic=..., attachments=..., chat_tier=...).
- For general tasks: Call
- Artifact Verification:
- Confirm downloaded files exist at the specified
<GENERATE_FILE>paths. - Verify file sizes and non-empty content.
- Confirm downloaded files exist at the specified
- Record in Walkthrough:
- Update
walkthrough.mdwith generated artifact paths and executive summary.
- Update
Completion criterion: All requested deliverables verified on disk and recorded in walkthrough.
Anti-Patterns
- Modality Over-Specification — Searching for 39 separate tools when
cellcog_runwith an explicit prompt andchat_modehandles all modalities seamlessly. - Credential Exposure — Wrapping
.envfiles, SSH keys, or secrets inside<SHOW_FILE>tags. - Tier Waste — Defaulting to
maxtier for simple icon or text generation thatflashhandles instantly. - Missing Output Tags — Omitting
<GENERATE_FILE>tags, resulting in downloaded files scattered across default cache folders. - Unbounded Delegation — Dispatching long-running cloud tasks without presenting the
implementation_plan.mdcheckpoint.
04 同仓库更多 Skills
ui-behavior-primitive-architect
Architect, audit, and verify zero-dependency headless UI behavior primitives, WebMCP agent bridges, WAI-ARIA/SSR/RTL tri-modal checks, and framework expert routing. Do not use for static CSS theme styling or backend database migrations.
trustgraph-ontology-governor
Govern knowledge graphs, SHACL-AF triple policies, OntologyRAG subclass validation, PROV-O dual named graphs, and Laya entropy decision scoring. Do not use for relational SQL queries or generic web scraping.
shim-harness-governor
Govern universal CLI agent loops, 4-tier tool permissions, ancestor-symlink workspace sandboxes, deterministic-first command routing, collision-safe MCP namespacing, and layered skill loading. Do not use for static UI styling or external SQL analytics.
graphiti-zep-memory-governor
Govern bi-temporal knowledge graphs with Graphiti future-edge clamping, MinHash/LSH entity dedup, RRF/MMR reranking, and Zep typed ontologies. Do not use for static SQL joins or generic web crawling.
agora-mcp-workbench-architect
Architect, audit, and orchestrate Agora Workbench MCP servers, loopback sidecars, unified state graphs, gateway/dispatcher connectors, and policy-enforced data lakes. Do not use for static frontend CSS styling or non-MCP desktop GUI automation.
skill-eval-assurance-runner
Evaluate and assure agent skills using small-N Wilson confidence intervals, unbiased pass^k reliability, deterministic trajectory graders, preflight boundary honesty, and sealed digest verification. Do not use for subjective prose review or unconstrained production deployment.