Unhobble

Audit and rightsize a context surface — CLAUDE.md, AGENTS.md, a SKILL.md, a system prompt, or a single prompt — against the Claude-5 context-engineering reversals. Measures the surface (token budget, how much of it is phrased as hard rules vs judgment, near-duplicate sections, contradiction candidates, filesystem-derivable content, progressive-disclosure gaps), then adjudicates each section keep / relocate / delete using a grounding-aware predicate: prose backed by an independent mechanism is decorative and free to cut, prose that is the only carrier of a behavior is a real bet. Also rewrites prompts and skills from prohibition-form into judgment-form. Anthropic deleted >80% of Claude Code's system prompt for Claude 5 with no eval loss; this is the repeatable version of that cut. Use when: (1) a CLAUDE.md or AGENTS.md has grown and you want to know what is safe to remove, (2) an agent is ignoring or colliding with its own instructions, (3) a prompt is long and over-constrained and underperforming, (4) auditing a skill for over-constraint before publishing, (5) onboarding a repo whose agent context was written for an older model. Triggers on unhobble, rightsize, trim my CLAUDE.md, audit my CLAUDE.md, is my CLAUDE.md too long, context engineering, prune governance, over-constrained, improve this prompt, my agent ignores instructions, conflicting instructions, prompt too long.

mkdir -p ~/.claude/skills/unhobble
curl -L https://claudskills.com/skills/unhobble/SKILL.md \
  -o ~/.claude/skills/unhobble/SKILL.md

@misc{unhobble-2026,
  author    = {broomva},
  title     = {Unhobble [Claude Code skill]},
  year      = {2026},
  publisher = {ClaudSkills},
  url       = {https://claudskills.com/skills/unhobble/}
}