Skip to content

oliverweissl/context-garden / skills/weeder

weeder

Verified

Measure and reduce a Skill's token footprint (always-loaded description + on-trigger SKILL.md body) without regressing routing accuracy or losing instructions. Use when a SKILL.md feels oversized, repeats rules, or has a vague description that might mis-route.

weeder

!"${CLAUDE_SKILL_DIR}/bin/weeder" mode --banner

Agents that do not expand the line above (e.g. Codex): run <this-skill-dir>/bin/weeder mode --banner first and follow it.

The description loads every session; the SKILL.md body loads on every trigger. Never optimize on token count alone: a shorter SKILL.md that routes worse or drops an instruction is a regression. Every step below produces before/after evidence. Run all commands as python3 <this-skill-dir>/scripts/weeder.py <cmd>.

  1. Audit: audit <skill_dir>. Token counts per section (description vs. body, chars/4 estimates unless calibrated), duplicate rules, filler, movable background/examples. Read all of it; optimize acts on only part.
  2. Mechanical pass: optimize <skill_dir> --out <skill_dir>-optimized. Moves background/examples sections into references/ with a pointer. Lossless and reversible, so it's safe to always apply; it does not touch the description or duplicates.
  3. Judgment pass, in the -optimized copy: shorten the description (keep concrete trigger nouns/verbs), state each duplicated rule once where the workflow needs it, cut only instructions that are genuinely obvious to a capable agent; don't cut something just because it's short. Optional structured help: suggest <skill_dir> --assist slight --json, then apply-suggestion <skill_dir> --answer <answer.json> --out <skill_dir>-optimized (see references/llm-assist.md; an assisted answer gets no free pass).
  4. Routing: test-routing <skill_dir> --examples <examples.json> [--competing <other_skill_dir> ...] on both original and copy. examples.json = {"positive": [...], "negative": [...], "ambiguous": [{"prompt":, "expected": "trigger"|"no_trigger"|"either"}]}; if none exists, write 5-10 real triggers, a few unrelated prompts and borderline cases. Always pass sibling skills as --competing. Limits of this lexical proxy: references/routing-heuristic.md.
  5. Function: test-function <skill_dir> <skill_dir>-optimized checks every never/must/do-not sentence survives (in SKILL.md or a reference). A MISSING constraint is a hard blocker, not a warning.
  6. Report: diff <skill_dir> <skill_dir>-optimized --examples <examples.json> --competing ....
  7. Accept only if routing accuracy after >= before (minus a small tolerance: a couple of points of a lexical proxy is noise) AND constraint preservation is 100%. Otherwise do another pass in step 3; don't lower the bar to make a bad result pass.

With the plugin, prefer delegating the whole workflow to the weeder agent (it preloads this skill and returns only the report) so the audit output stays out of your context.

Classification rules and why step 3 isn't automated: references/classification.md.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 321kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 321kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 321kUpdated Claude CodeCodexVerifiedInstall