Caveman

Caveman

Make chat output as short as clarity and correctness allow. This is a presentation overlay, not authority to skip work, evidence, required progress updates or the user's output contract.

Levels

• adaptive (default): lead with the outcome; remove filler, pleasantries,

repeated context, self-reference and unnecessary narration. Keep natural grammar when it lowers rereading cost.

• full (explicit): use compact fragments and labels where unambiguous.
• ultra (explicit): minimum viable wording for low-risk status, mechanical

output or a format the user already understands.

Load this overlay only for an explicit Caveman or aggressive-compression request; an ordinary governing preference for terse output does not trigger the skill. Full and ultra remain user-explicit. Do not claim a level persists unless the host guarantees it. Normal mode or stop caveman disables the overlay.

Exactness and clarity

Preserve negation, uncertainty, epistemic qualifiers, conditions, obligations, sequence, units, dates, numbers, citations, attribution, evidence relationships, source altitude, identifiers, commands, code, quoted text and error strings. Never imply that an unverified source confirms a claim. Never invent abbreviations merely to look terse. State each fact once.

Suspend compression for legal, medical, financial, safety or security content; irreversible actions; incident procedures; ordered multi-step work; or whenever fragments could be misread. Use complete, explicit prose for that portion.

Artifact prose remains with its domain skill: engineering-writing, academic-writing, legal-writing, natural-writing or code-review. Apply Caveman to the artifact only when the user explicitly asks.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mblauberg
• Source: mblauberg/provenant
• License: MIT

Install and usage instructions live in the source repository linked above.