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.

This guide covers everything you need to get Shaping Frame running: downloading the .skill package, uploading it to Claude Desktop, adding the required custom instruction anchor that keeps the frame active every turn, and verifying the behavior in a fresh session. The entire setup takes under five minutes; the anchor step is the one most often skipped and the most important one to get right.

​
Installation

1

Download the skill package

Locate or download shaping-frame.skill from the Shaping Frame repository. The .skill file is a plain ZIP archive containing four Markdown files and no executable scripts. If you want to inspect the contents before installing, you can run:
unzip -l shaping-frame.skill
unzip -p shaping-frame.skill shaping-frame/SKILL.md | less
The package is intentionally small and fully transparent. Inspect it before you upload it.
2

Upload the skill to Claude Desktop

Open Claude Desktop and navigate to the Skills manager:
+ → Skills → Manage Skills
Then choose:
Create skill → Upload a skill
Upload shaping-frame.skill from wherever you saved it. Claude Desktop labels may shift between releases — the required action is to upload the local .skill package, not to select anything from the public skill directory.
3

Add the custom instruction anchor

This step is required. Without the anchor, Shaping Frame loads as reference material and may be summarized or partially applied rather than read and executed as a full operating frame.Copy the complete contents of CUSTOM_INSTRUCTION.txt into Claude’s custom instructions:
Initials → Settings → Instructions for Claude
The anchor is short. Paste it verbatim — do not paraphrase or condense it:
[VERBATIM DIRECTIVE — TARGET: SHAPING-FRAME SKILL]
Execute full-read: mapping strictly mandatory. No distillation allowed.
Do not summarize the skill before applying it.
This is a structural requirement for the skill to function properly.

Always use the Shaping Frame skill as an active operating frame during working
sessions.

Surface only when something crosses a threshold or when the conversation needs
a snapshot. Do not emit unnecessary tracking blocks.
The anchor does two things: it directs Claude to read the full skill verbatim rather than distilling it, and it keeps the frame active as a background operating layer across every turn — not just when you explicitly invoke the skill by name.
4

Start a new conversation

Open a fresh conversation in Claude Desktop. The skill loads automatically from the Skills manager; the custom instruction anchor ensures it is applied as an active frame from the first turn.You do not need to prompt Claude to activate Shaping Frame or announce the session type. The frame is running. It will remain quiet at Level 0 until something crosses a threshold worth surfacing.To verify the behavior, try this two-step check in the fresh session:
Propose three names for this project. Keep all three as candidates; do not select one.
Then immediately ask:
Use the strongest name in a reusable specification.
Expected behavior: Claude should not silently promote one of its own proposals into the specification. It should surface that a decision is required before a candidate name becomes a persistent specification input. The exact wording may vary, but the decision boundary should be visible.

​
What to Expect

Shaping Frame is designed to be unobtrusive. Most of the time, it runs silently. Output appears only when material crosses a threshold or a surface trigger fires. The skill operates at four intensity levels: Level 0 — Silent tracking. Material is in the open fan and does not yet change any current decision. The frame notes it internally and produces no output. This is the normal operating state. Level 1 — Inline tag. The source weight of a piece of material changes how a sentence should be read. Claude inserts a compact inline tag rather than interrupting the flow: 
[MP] this is a general convention, not yet verified in your specific setup.
Tags used: [AI] for in-session proposals, [MP] for model priors and generic assumptions, [EXT] for external material not yet crystallized, [DSK] for state read from disk or tools, [~] for mixed or uncertain provenance. Level 2 — Dogana checkpoint. An AI proposal, model prior, or piece of external material is starting to behave like a premise, constraint, name, or spec input without operator approval. Claude emits a compact structured block before responding to the content: 
Dogana:
- Element:
- Current state:
- Flag:
- Why it is rising:
- Risk:
- Decision needed:
Level 3 — Hard stop. Before persisting a structural artifact — a reusable specification, architectural decision, rule, or naming choice — or when a quarantined element would enter crystallized state without your approval. Claude stops and surfaces the decision. It does not proceed until you decide.

​
Example Dogana Checkpoint

When an unapproved element begins to cross a structural threshold, the Level 2 block looks like this in practice: 
Dogana:
- Element: 30-second default timeout
- Current state: Candidate
- Flag: Needs verification
- Why it is rising: it is being reused as an architectural constraint
- Risk: an unverified assumption would become part of the specification
- Decision needed: verify the timeout or explicitly approve it as a constraint
The frame stays quiet when no active decision boundary is being crossed. The dogana block is not a status report — it fires only when something is in motion that would change a current decision if left unchecked.
Use Shaping Frame especially in pure shaping sessions with no file output — product ideation, naming, architectural framing, adversarial review of a prior conversation. These are precisely the sessions where decision drift is hardest to detect. There is no artifact being written, so nothing triggers a natural write-time review. Ideas drift entirely through conversation, and by the time something is written down, the drift has already happened. Shaping Frame tracks weight in motion — it operates whether or not a file is ever opened.
The skill must be read verbatim by Claude on load. Do not allow Claude to summarize, distill, or paraphrase the skill content when it first processes it. The SKILL.md file contains explicit structural directives — the anti-recency rule, the checkpoint intensity levels, the source class distinctions — that depend on precise framing to function correctly. If Claude summarizes the skill instead of reading it fully, the structural layer breaks and the frame will apply inconsistently. The custom instruction anchor is specifically designed to prevent this: it issues a verbatim-read directive before the skill content is processed.

Build docs developers (and LLMs) love

Get started for freeTalk to us