Skip to content

Add memento-style compact context artifacts for token-efficient task projections #924

Description

@tcconnally

Source motivation

Memento and recent context-engineering research emphasize compact, reusable intermediate context artifacts that preserve future-use signal while materially reducing context cost.

Problem

Perseus can assemble rich context, but it does not yet treat compact derivative context artifacts as a first-class product surface.

Proposal

Add memento-style context compaction artifacts to Perseus.

Potential scope:

  • introduce a compact artifact tier distinct from full context and narrative summary
  • preserve task objective, constraints, unresolved questions, evidence anchors, and next-step recommendations
  • expose token-budget-aware generation and evaluation paths
  • benchmark compact artifacts against full context on representative tasks

Why now

Recent research shows context compaction is not just an optimization detail. It can materially improve cost-efficiency and downstream reasoning quality when done well.

Success criteria

  • compact artifacts consistently smaller than full task context
  • no major regression in benchmark task quality
  • measurable reduction in context budget consumption

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions