Dream Persona Distillation and Relationship Graph

Problem It Solves

In character-centric workflows, persona profiles, relationship graphs, and dialogue context often live in scattered prompts, ad-hoc scripts, and temporary files. A host LLM must repeatedly assemble inputs, check whether artifacts are complete, and handle long-form truncation or incremental updates. zaomeng-skill turns these steps into standard payloads, artifact statuses, and a runtime index that a host can drive programmatically.

How It Works

The skill follows an LLM-first / prompt-first model: the host calls the model and advances the conversation, while the skill builds standard requests, validates outputs, maintains run_manifest.json, and exposes helper tools.

The core pipeline includes:
- distill: builds persona distillation payloads, supports incremental updates and long-text chunking, and targets PROFILE.generated.md.
- materialize: turns generated output into a persona bundle directory and writes ARTIFACT_STATUS.generated.json.
- export_graph: exports relationship graphs in HTML, Mermaid, SVG, and related formats.
- verify_workflow: validates that the full workflow produced the expected artifacts.
- dialogue helpers: self cards, persona field autofill, act / insert / observe reply suggestions, and next-scene recommendations.

When text is too long, the payload returns chunks[] and merge_payload, letting the host generate partial drafts per chunk before merging. In incremental mode, existing profiles are added to request.existing_profiles, and run_manifest.json records update_mode plus the matched character count.

Boundaries

This is not a standalone chat CLI and does not include model invocation. It assumes the host can run the Python tools, read JSON status files, and progress through run_manifest.json stages. It fits projects that want persona distillation, bundle materialization, graph export, and dialogue assistance to be inspectable and repeatable, not lightweight prompt templates that skip artifact state.