curt-park/triage-knowledge

Overview

Scan docs/ against conversation context or an action summary. Classify each knowledge item as existing or new, and return topology hints (connected nodes, shared tags) so sync and capture can use them during execution.

Process

1. Extract knowledge items from context

From conversation or action summary, identify:

• Decisions made, conventions established, components changed, patterns discovered

2. Search docs/ for each item

For each item:

• Grep docs/ for keywords, titles, tags
• Read matched files to confirm relevance

3. Build topology hints

For matched nodes:

• Extract wikilink connections ([[target]] patterns)
• Identify nodes sharing tags
• Flag hub/orphan status

For unmatched new items:

• Suggest related existing nodes based on tags or content overlap

4. Return classified list

Output Format

Autology — Triage Results

Existing (→ sync)

• docs/foo.md — matches [item description] Connected: [[bar]], [[baz]] | Tags: arch, api | Hub (5 links)
• docs/qux.md — matches [item description] Connected: [[foo]] | Tags: convention | Orphan-adjacent

New (→ capture)

• [item description] — no matching node Suggested relations: [[foo]], [[bar]] (shared tags: architecture)
• [item description] — no matching node Suggested relations: none

If no existing docs match any items, use empty-state format:

Autology — Triage Results

No existing matches found. All items classified as new.

New (→ capture)

• [item description] — no matching node Suggested relations: [[foo]] (shared tags: architecture)

Common Mistakes

| Mistake | Fix | |---------|-----| | Judge relevance by keyword match alone | Read the file to confirm actual relevance | | Omit topology hints | sync/capture rely on connected/suggested relations — always include | | Ignore implicit relations | Check tags and content overlap even without wikilinks |