Menü
Use when a skill SKILL.md is over 2000 words and needs trimming to the Superpowers standard of under 500 words.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Use when Hermes Agent needs X/Twitter search, public signal research, monitoring, or approval-gated account actions through the Hermes Tweet plugin.
Router Learning System — 越用越聰明嘅任務路由引擎。 每個 task 有最佳 path，Router 自動記錄經驗、動態調整、跳過爛 path。
Use when signaling work status across autonomous sessions.
Use when needing to standardize descriptions across many skill files at once — change format, fix wording, or enforce a convention like 'Use when...' prefix.
Use before making irreversible changes (deleting files, destructive git operations, large refactors). Use when you need the ability to roll back if something goes wrong. Use at the start of any task that modifies multiple files.
Use when the context window is getting full and tasks need breaking down.
| name | skill-slimming-strategy |
| description | Use when a skill SKILL.md is over 2000 words and needs trimming to the Superpowers standard of under 500 words. |
| metadata | {"hermes":{"platform":["cursor","codex","claude-code"]}} |
| domain | ecosystem |
| subdomain | agent-skills |
| tokens | {"scan":105,"load":1200,"category":"standard"} |
When a SKILL.md file exceeds 2000 words and needs to be made more token-efficient per Superpowers standards.
Before slimming, classify the skill into one of these types. Each gets a different treatment:
Signs: Language like "Created On: Jun 06, 2025 | Last Updated On:..." or "This skill auto-generated from official documentation" Action: Replace entirely. Keep only the frontmatter + a 200-word summary with links to official docs. The original content is stale anyway — official docs are more current.
Example:
# Before: 22,733 words of copied PyTorch API docs
# After: 219 words — frontmatter + 3 bullet core concepts + 3 common tasks + links
Signs: Has step-by-step phases, checklists, decision trees. Written by a human for a specific project.
Action: Extract details to reference files. Keep the decision tree / flow diagram + one-sentence-per-phase in SKILL.md. Move detailed checklists, commands, and troubleshooting to references/ files.
Example:
# Before: 14,058 words — full paper writing pipeline with all 7 phases detailed
# After: 429 words — flow diagram + phase summary table + links to reference files
Signs: Has homepage: or repository: pointing to an external GitHub project. Contains vendored Python scripts.
Action: Don't touch. These are maintained by external authors. The scripts are the real content, not the SKILL.md prose.
Example: last30days — 19,231 words of scripts + instructions from mvanhorn.
Signs: Contains real-world debugging steps learned through painful experience. Lots of "quick fix" ordering.
Action: Abstract to decision tree. Keep the symptom diagnosis (Case A vs Case B) + quick-fix order in SKILL.md. Move deep-dive explanations and repair procedures to references/.
Example:
# Before: 8,022 words — full Cocos gray screen debug flow with 5 case analyses
# After: 209 words — symptom check + quick-fix order + reference links
Understand the structure: is it a flow, a reference, or an external package?
Use the classification above.
references/ files, keep outline onlywc -w skills/path/SKILL.md # Target: <500 words
homepage: or repository: in frontmatter before slimming.references/ files so it's still accessible via the Read tool when needed.claude-code, hermes-agent, and project-specific debugs are frequently loaded. Slimming them requires care.batch-skill-description-standardization for the batch approach pattern).