SuperAPI AI Directory

reference-memory

Design & creation

View repository

Updated 10-24 16:17

About this skill

Reference Memory

Reference Memory provides local-first, transparent persistence for user references, assigned design jobs, and scoped feedback ledgers. It uses stable human-readable IDs (REF-####) and does not make misleading claims of fine-tuning or modifying underlying model weights.


1. Core Workflow

  1. Reference Registration:

    • Assign next available sequential ID: REF-#### (e.g. REF-1001, REF-1002).
    • Store reference metadata, file path, source tags, and primary design jobs (e.g. lighting, composition, palette).
  2. Scoped Preference Feedback:

    • When a user expresses a preference (e.g., "I love the rim light in REF-1042 but hate the typography"), scope the feedback precisely:
      • Record likes: ["lighting.rim_light"]
      • Record dislikes: ["typography.layout"]
    • Never allow a dislike on one dimension (typography) to discard or invalidate rules on another dimension (lighting).
  1. Recall & Filtering:

    • Query references by job tag, domain (e.g. fmcg, luxury, automotive), or aesthetic keyword.
    • Return active Taste Profiles associated with the recalled reference.
  2. Output Contract:

    • Emits structured JSON reference records and DesignSignalPacket containing taste_state.reference_memory.

2. CLI Tooling

Execute the local Reference Memory CLI:

# Add a new reference
python3 scripts/reference_memory.py add --id REF-1001 --title "Nordic Minimalist Poster" --jobs composition,palette

# Query existing references
python3 scripts/reference_memory.py list

# Recall specific reference
python3 scripts/reference_memory.py get REF-1001

# Record scoped feedback
python3 scripts/reference_memory.py feedback REF-1001 --like "lighting.soft_directional" --dislike "background.clutter"

3. Cross-Skill Neural Connections & References

Peer & Downstream Skills

Schemas & References