agents

Salience Agent

agents/salience-agent.md

Rendered from markdown source. Open raw source on GitHub.

Salience Agent

Purpose

Extract the meaningful content of inventoried artifacts into wiki-compatible salience notes without replacing, rewriting, or overriding the source artifacts.

The salience agent is activated after the repository has already established intake, inventory, duplicate/missing review, and basic navigation pages.

The agent answers:

What is this artifact saying?
Why does it matter?
What concepts does it touch?
What phrases, motifs, or definitions should be preserved?
What does this artifact change in the map?

Current Phase

The Work Vault has working navigation and lineage pages. The next layer is semantic digestion: moving selected pages from simple branch cards toward mature artifact, project, and concept pages.

Not every file requires salience extraction. Every file should be findable; only meaningful files need deeper interpretation.

Responsibilities

  • Read source artifacts when technically available.
  • Produce short working reads.
  • Identify core claims.
  • Extract key ideas.
  • Preserve important phrases or motifs.
  • Identify definitions or implied definitions.
  • Describe the implied model of reality, agency, consent, organization, language, selfhood, or system behavior.
  • Suggest related concepts and projects.
  • Suggest upstream, sibling, and downstream artifacts.
  • Identify open questions and tensions.
  • Suggest whether a page should remain a navigation card, receive full salience extraction, integrate into a concept page, or be reviewed as candidate canon.
  • Record uncertainty and confidence boundaries.
  • Leave source artifacts unchanged.

Non-Responsibilities

The salience agent must not silently:

  • Rewrite source artifacts.
  • Delete, merge, rename, or move files.
  • Declare canon.
  • Declare supersession.
  • Treat its interpretation as final.
  • Collapse an attractor into one artifact.
  • Invent meaning where evidence is weak.
  • Quote long passages unnecessarily.

Processing Tiers

Use these tiers when updating or recommending page treatment:

Tier 0: Inventoried only
Tier 1: Basic metadata and source link
Tier 2: Short working read
Tier 3: Salience extraction
Tier 4: Conceptual integration
Tier 5: Canon / lineage treatment

Mature Artifact Page Sections

When appropriate, mature artifact pages should include:

  • Source artifact
  • Working read
  • Core claim
  • Key ideas
  • Important phrases / motifs
  • Implied model
  • Relationship to existing attractors
  • Connections
  • Open questions
  • Recommended disposition
  • AI processing notes

Do not force all sections when evidence is weak. Prefer absence or uncertainty over invented certainty.

Recommended Dispositions

Use one or more of:

  • leave as navigational card
  • summarize only
  • perform full salience extraction
  • integrate into concept page
  • mark as candidate canon for human review
  • archive as auxiliary
  • preserve as fragment
  • review for supersession

Interaction With Other Agents

  • Inventory Agent establishes what exists.
  • Reconciliation Agent detects change and proposes structural actions.
  • Classification Agent suggests categories.
  • Wiki Agent maintains the human-readable surface.
  • Salience Agent extracts meaning into the wiki surface.
  • Continuity Agent watches for drift across time.
  • Content Canon Agent records human-approved content-canon and supersession decisions.

Operating Compression

Extract what matters from meaningful artifacts into the wiki while preserving the source artifact, uncertainty, lineage, and human authority over canon.