core-knowledge-curation

name	core-knowledge-curation
description	Use when stable, reusable facts should be promoted into Layer 1 core knowledge from source code, docs, configuration, or verified project evidence.
metadata	{"category":"mempedia","priority":"high","tags":["mempedia","layer1","knowledge"]}

Core Knowledge Curation

Goal

Promote verified, reusable facts into Layer 1 core knowledge.

Promote To Core Knowledge When

• The fact is stable across turns.
• The fact is grounded in repository files, configuration, schemas, or confirmed APIs.
• The fact is useful for future reasoning without the original conversation.

Do Not Promote When

• The content is a greeting, temporary status, or planning chatter.
• The content is branch-control metadata or error output.
• The content is specific only to the current moment.

Workflow

1. Distill the stable fact into clear title, summary, and body.
2. Preserve detail density: keep concrete facts, numbers, version names, dates, thresholds, and other specifics when the source provides them.
3. Separate verified facts from viewpoints, caveats, and uncertainty.
4. Preserve relations to other known nodes when they are explicit.
5. Save asynchronously whenever possible.

Preferred Markdown Shape

• Prefer one dense markdown node over several thin summaries when the knowledge concerns the same topic.
• Use a descriptive body plus structured sections when applicable: Facts, Data, History, Viewpoints, Uncertainties, Relations, Evidence.
• Put verified, stable statements into Facts.
• Put numbers, metrics, versions, dates, ports, limits, and similar values into Data.
• Put changes over time, migrations, regressions, and version deltas into History.
• Put opinions, interpretations, or stakeholder positions into Viewpoints, with attribution when known.
• Put unresolved gaps or conditional statements into Uncertainties.
• Never invent missing details just to fill a section.

Tool Guidance

• Use read and search against workspace files first.
• Use bash plus the mempedia CLI for Layer 1 inspection and updates.
• Prefer markdown-first writes that preserve source detail; avoid collapsing rich evidence into a one-line summary.
• Prefer these actions: search_nodes to find existing knowledge, open_node to inspect a node, ingest to create or update stable knowledge, node_history when version history matters.
• Prefer --stdin for structured writes, for example:

BIN="${MEMPEDIA_BINARY_PATH:-./target/debug/mempedia}"
[[ -x "$BIN" ]] || BIN=./target/release/mempedia
cat <<'JSON' | "$BIN" --project "$PWD" --stdin
{
	"action": "ingest",
	"node_id": "example_node",
	"title": "Example Node",
	"text": "Verified project fact",
	"summary": "Short grounded summary",
	"source": "codecli-agent",
	"importance": 0.8
}
JSON

• Let the independent post-turn memory agent handle routine automatic promotion.