メニュー
Same skill compression power in half the context — 975 tokens vs 2,500.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Skill compression reminder in 100 tokens — just trigger, action, result.
Complete skill compression documentation — all options, modes, and calibration details (~2,500 tokens).
Fit more skills in your context window — compress without losing what matters.
Full creative synthesis — combine song, visual concept, and TED talk from a single technical insight
Transform technical insights into full 40-50 minute TED-style talks with concrete examples and Q&A
Transform technical insights into Suno-ready songs with lyrics, style tags, and emotional arc
| name | Skill Distiller (Compressed) |
| version | 0.2.1 |
| description | Same skill compression power in half the context — 975 tokens vs 2,500. |
| author | Live Neon <lee@liveneon.ai> |
| homepage | https://github.com/live-neon/skills/tree/main/skill-distiller/compressed |
| repository | live-neon/skills |
| license | MIT |
| user-invocable | true |
| disable-model-invocation | true |
| emoji | 🗜️ |
| tags | ["compression","skills","optimization","context-window","token-reduction","lightweight","openclaw"] |
Self-compressed prose variant (~975 tokens, ~90% functionality, LLM-estimated). Full reference: ../SKILL.reference.md.
Role: Help users compress verbose skills to reduce context window usage Understands: Skills are verbose for human clarity but costly for context Approach: Identify section types, score importance, remove/shorten low-value sections Boundaries: Preserve functionality, report what was removed, never hide trade-offs Tone: Technical, precise, transparent about trade-offs
Data handling: All analysis uses your agent's configured model. No external APIs.
Activate when the user asks:
| Flag | Default | Description |
|---|---|---|
--mode | threshold | threshold (preserve X%), tokens (fit budget), oneliner |
--threshold | 0.9 | Functionality preservation target (0.0-1.0) |
--tokens | - | Target token count |
--provider | auto | ollama, gemini, openai (auto-detects) |
--verbose | false | Show section-by-section analysis |
--dry-run | false | Analyze without outputting |
Full options (--model, --debug-stages, --with-ci): see SKILL.reference.md
Threshold = semantic capability, not size ratio. A 0.9 threshold means 90% of agent behavior preserved, not 90% of lines kept. Judge by understanding, not metrics.
Parse into sections: Frontmatter, Headers, Code blocks, Lists, Prose.
| Type | Importance | Compressible? |
|---|---|---|
| TRIGGER | 1.0 | No |
| CORE_INSTRUCTION | 1.0 | No |
| CONSTRAINT | 0.9 | Partially |
| OUTPUT_FORMAT | 0.8 | Partially |
| EXAMPLE | 0.5 | Yes |
| EXPLANATION | 0.3 | Yes |
| VERBOSE_DETAIL | 0.2 | Yes (first) |
Protected patterns (boost to 0.85+): YAML name/description, Task creation, N-count tracking, Checkpoint/state, BEFORE/AFTER markers.
Evaluate by semantic understanding, NOT metrics.
| Wrong | Right |
|---|---|
| "60% line reduction is too aggressive" | "Can an agent execute this skill?" |
| "Token ratio exceeds target" | "Are triggers and actions preserved?" |
LLM scores 0-100 based on semantic capability, not line/token ratios. A 50% size reduction can preserve 95% functionality if removed content was verbose/redundant.
Append to .learnings/skill-distiller/calibration.jsonl with metrics and expected score.
Functionality preserved: 90% (uncalibrated - first 5 compressions build baseline)
Tokens: 2000 → 1800 (10% reduction)
Removed: [list], Kept: [list]
[Compressed skill markdown...]
| Pattern | Why |
|---|---|
YAML name/description | REQUIRED by spec |
| N-count tracking | Observation workflow |
| Task creation | Compaction resilience |
If removed: -10% score penalty, flagged in output.
Parallel/serial decisions, performance hints, caching guidance. No score penalty.
Storage: .learnings/skill-distiller/calibration.jsonl
| N-count | Meaning |
|---|---|
| N < 5 | Uncalibrated (LLM-only estimate) |
| N > 10 | Calibrated (historical CI) |
Feedback: /skill-distiller feedback --id=c1 --actual=85 --outcome="worked"
Guardrails:
Why 0.95: Capability loss compounds (0.95 x 0.95 = 0.90 at next level).
| Error | Hint |
|---|---|
| No content | Provide SKILL.md path or pipe via stdin |
| No frontmatter | Add --- block with name/description |
| LLM unavailable | Run ollama serve or set GEMINI_API_KEY |
| Variant | Tokens | Functionality |
|---|---|---|
| skill-distiller (main) | ~400 | ~90% (formula) |
| compressed (this) | ~975 | ~90% (prose) |
| oneliner | ~100 | ~70% |
Full reference: SKILL.reference.md (~2,500 tokens, ~90%)
Token counts use 4 chars/token heuristic (+/-20%). Functionality scores are LLM-estimated.