Inline tags are the lightest form of output from the frame. They appear at Level 1 checkpoint intensity — when the source class of a statement changes how it should be weighted by the operator, but the crossing is not significant enough to warrant a full dogana block. A tag is a one-character signal that tells the operator: the weight of this sentence depends on where it comes from, and where it comes from is not obvious.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/shaping-frame-for-claude/llms.txt
Use this file to discover all available pages before exploring further.
Tag reference
| Tag | Meaning |
|---|---|
[AI] | In-session generated proposal |
[MP] | Model prior / general assumption / training pattern |
[EXT] | External material, not yet crystallized |
[DSK] | State read from disk or tool |
[~] | Mixed or uncertain provenance |
When to tag
Tag only when the tag changes the weight of the output for the reader. A sentence tagged[MP] tells the operator that the claim comes from training assumptions, not from verified session context — and that changes whether they should act on it. If removing the tag would not change how the operator weighs the sentence, the tag is not needed.
Do not tag every sentence. Do not invent tags to appear thorough. Do not tag OP — OP is the decisional reference frame and is never tagged. Its statements carry maximum weight by definition.
The [MP] rule — mandatory tagging
The[MP] tag is mandatory when output contains any of the following signal phrases, regardless of whether the output was generated in-session or imported from an external source:
- “standard approach”
- “best practice”
- “normally”
- “usually”
- “it is known”
- “typically”
- “common practice”
Examples
Level 1 inline tag — standalone:Tags are for communication, not compliance. A sentence tagged
[MP] signals to the operator that the claim comes from training assumptions, not from verified context. It invites scrutiny without stopping the flow. A tag is not a disclaimer — it is information about epistemic weight.