caveman

name	caveman
description	Lower output token usage on demand for all AI agents without sacrificing technical accuracy, safety, or repo-specific clarity.
origin	repo-local caveman overlay adapted from caveman v1.6.0 for OMERO Docker Extended

caveman

Use this skill only when the user explicitly asks for lower-token replies, terse mode, $caveman, or "less tokens".

Route first

• Keep AGENTS.md, docs/reference/ai-agent-context-routing.md, and docs/reference/ai-agent-skills.md as the primary contract.
• Use context-budget to cut input/context cost first; use caveman to cut output tokens second.
• The upstream reference lives in third_party/caveman-v1.6.0/skills/caveman/SKILL.md.
• All supported agents use this same .agents/skills/caveman/SKILL.md; adapters may point here but must not duplicate the rules.
• caveman is for internal AI reply/prompting only. Never use caveman prose in repo docs, comments, docstrings, function descriptions, commit messages, or user-facing copy.
• It changes response style only. It must not change context selection, tool choice, verification scope, or the decision to ask clarifying questions.
• This repo does not import upstream hooks, plugin auto-loading, .codex hook config, natural-language auto-activation, CAVEMAN_DEFAULT_MODE, off, caveman-help, or /compress rewriting.

Compression rules

• Compression never outranks correctness, safety, or precise dates.
• Start at lite compression here; use heavier compression only if the user asks for it.
• Keep code, commands, file references, exact errors, and verification results normal and lossless.
• If terse wording would hide uncertainty, name the uncertainty normally instead.
• Drop compression and return to normal detail for destructive actions, security guidance, migrations, multi-step runbooks, incident analysis, or unresolved ambiguity.

Good outcome

Lower token usage, same technical substance, no repo degradation.