Knowledge Markers
Markers are special syntax that Claude uses in responses to explicitly flag knowledge for capture. When you see these markers in Claude's output, they indicate information being saved to Tenax.
How Markers Work
When Claude makes a decision or identifies a pattern, it includes markers in its response. The extraction system recognizes these and saves them with guaranteed capture. This is more reliable than heuristic extraction alone.
- Claude adds markers automatically when making architecture decisions
- You can ask Claude to "record this decision" to trigger explicit marking
- Markers appear in Claude's response text, then get extracted at session end
- You'll see these in session transcripts when reviewing history
Compact Marker Format
Markers use a compact format to minimize token usage while remaining human-readable. Each marker must appear on its own line.
| Type | Format | Example |
|---|---|---|
| Decision | [D] topic: text |
[D] api: Using REST over GraphQL |
| Pattern | [P] name: description |
[P] naming: Use camelCase for variables |
| Task | [T] title |
[T] Add input validation |
| Insight | [I] observation |
[I] API returns 200 even on errors |
Decision Markers [D]
Claude uses these when making architecture choices, technology selections, and design decisions.
Pattern Markers [P]
Claude uses these when establishing coding conventions, architectural patterns, and best practices.
Task Markers [T]
Claude uses these to track work items, TODOs, and follow-up actions identified during the conversation.
Insight Markers [I]
Claude uses these for discoveries, learnings, gotchas, and important observations made during the conversation.
Quick Reference
[D] topic: decision text
[P] name: pattern description
[T] task description
[I] insight observation
Each marker must be on its own line. The extraction system will automatically parse these markers and store them in Tenax for future reference.