Internal Lang

Internal Lang

Purpose

Provide an explicit, conversation-local switch for compact notation without changing normal task behavior.

Remain inactive unless the user invokes a supported command or directly requests this notation. Activation does not authorize repository changes or durable note creation.

State Model

Before explicit activation, assume no internal-lang behavior.

On the first explicit activation in a conversation, initialize:

1. internal=true
2. response=false

Then apply commands in message order:

1. --internal-lang on sets internal=true.
2. --internal-lang off sets internal=false and leaves response unchanged.
3. --internal-lang --response on sets response=true and does not force internal=true.
4. --internal-lang --response off sets response=false.
5. The newest explicit command wins for the state it changes.
6. Ignore malformed or unsupported command tokens and briefly identify what was ignored.

Do not assume state persists into a new conversation unless the runtime explicitly preserves it.

Response Boundary

When response=false:

1. keep user-facing language normal
2. omit compact scratch notation from the final response
3. mention state only when it changes

When response=true:

1. use compact notation sparingly
2. provide a key when meaning may be ambiguous
3. expand safety-critical, user-actionable, or validation-critical content into normal language

Safety Gate

Never use compact notation to obscure or lossy-compress:

1. user requirements or acceptance criteria
2. legal, security, financial, or credential material
3. commands, code, schemas, or exact error messages
4. blockers, uncertainty, or validation evidence

Do not expose hidden chain of thought or claim that this switch changes the model's internal neural representation.

Conditional Reference

Read Notation and Safety only when compact notation will actually be used, expanded, or saved in an explicitly requested artifact.

Completion Check

1. Every explicit command produced the documented state transition.
2. Normal responses remained normal while response=false.
3. Compact content remained reversible and non-lossy.
4. No file was created or changed solely because this skill activated.