Skip to main content

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.

The shaping-frame.skill file is a ZIP archive uploaded to Claude Desktop through the Skills panel. Inside the archive are four files that work together as a layered cognitive frame: a full operating specification, a compact reminder, a checkpoint checklist, and a surface snapshot format. Claude reads each layer at a different point in the conversation, keeping the full frame active while only loading heavier references when they are needed.

​
Package structure

shaping-frame.skill
└── shaping-frame/
    ├── SKILL.md
    └── references/
        ├── frame.md
        ├── dogana.md
        └── snapshot.md

​
The layer model

The frame operates across five layers. Four of them are files inside the archive. The fifth — the floor — is the custom instruction anchored in userPreferences, which keeps the frame active every turn outside of any file read. Each layer is read at a specific moment for a specific purpose.
LayerFileRoleWhen read
Flooranchor in userPreferencesAlways in context every turnEvery turn, without exception
Full frameSKILL.mdComplete operating specification. Verbatim read required.Active for the entire conversation
Compact reminderreferences/frame.mdSummary of sources, states, and threshold rulesAt session start and on drift risk
Checkpoint checklistreferences/dogana.mdStep-by-step checklist for threshold crossingsWhen a checkpoint threshold fires
Surface snapshotreferences/snapshot.mdFormat template for active tracking blocksWhen a surface trigger fires

​
SKILL.md — the full frame

SKILL.md is the complete operating specification for the skill. It defines sources, weight states, thresholds, surface triggers, checkpoint intensity levels, inline tagging rules, the pre-write gate, and decontamination behavior. The skill header carries an explicit verbatim read requirement: 
Execute full-read: mapping strictly mandatory. No distillation allowed.
Read verbatim: disable semantic summarization for all the "shaping frame"
skill or related components. This is a structural requirement for the skill
to function properly.
This means Claude must not summarize or compress the skill before applying it. The full specification must be mapped into the operating frame, not distilled into a shorter approximation. This behavior is reinforced and anchored by the custom instruction — without the CUSTOM_INSTRUCTION.txt anchor in place, this requirement may not be respected.

​
references/frame.md — the compact reminder

references/frame.md is a condensed version of the core rules: source classes, default weights, weight states, threshold vocabulary, surface trigger conditions, the anti-recency rule, and checkpoint intensity levels. Claude reads frame.md at the start of each working session to orient the frame before the conversation moves. It reads it again if drift is detected — meaning material has started behaving as if decided without OP approval. The compact reminder is designed to be fast to re-read and re-apply without interrupting the session.

​
references/dogana.md — the checkpoint checklist

references/dogana.md contains the full checklist of controls to run when a checkpoint threshold fires. It lists all nine threshold crossings, twelve diagnostic controls, the dogana block output format, a four-level intensity guide, and the quick-reference tag table. Claude reads this file when it detects that material is crossing a threshold — for example, when an AI-generated proposal starts acting as a premise for a structural decision, or when a model prior is about to ground an architecture choice. The checklist makes the crossing visible and provides the format for surfacing it to the operator.

​
references/snapshot.md — the surface snapshot format

references/snapshot.md defines the format of the active tracking block that Claude emits when a surface trigger fires. It specifies the five fields of each tracked element, the states that qualify for inclusion, the states that do not, and the ordering rule for output. Claude reads this file when a surface trigger occurs — for instance, when the operator changes direction, asks for orientation, or opens a new artifact. The snapshot block is emitted before Claude responds to content, giving the operator a view of what is in motion before the conversation advances.
The CUSTOM_INSTRUCTION.txt file is not part of the .skill archive. It must be added separately to Instructions for Claude in Claude Desktop settings.

Build docs developers (and LLMs) love

Get started for freeTalk to us