EfficientSkill: Learning Skills for Token-Efficient Multi-Turn Agents
Organizations: Google LLC · Purdue University
Abstract
Reflective skill optimizers improve multi-turn LLM agents by revising instructions from failed and successful rollouts, yet accumulating edge-case rules bloats prompts, inflates token use, and pushes agents into long, error-prone loops. We introduce EfficientSkill, a two-step framework that first discovers a reliable skill without token penalties, and then warm-starts accuracy-guarded pruning from that skill to compress instructions, bound agent turns, and offload reusable code into workspace utilities with zero implementation prompt cost. Across four frontier, mid-tier, and open-weight models on MATH-500, ALFWorld, and SpreadsheetBench, EfficientSkill cuts output tokens over unconstrained EvoSkill on 11 of 12 configurations (by up to 80%) while raising mean accuracy on every benchmark, including +14.3 and +17.5 percentage-point gains on SpreadsheetBench with Gemini 3.8 Flash and Grok 4.1 Fast by replacing stalled loops with compact workflows. By shrinking both generated tokens and replayed history, EfficientSkill lowers total token usage and serving cost across every evaluated configuration, with or without full-prefix KV caching.
Figures & tables
| Algorithm 1: EfficientSkill |
|---|
| Input: Agent , seed , , |
| Output: Token-efficient skill |
| Phase 1: Discover Anchor // unconstrained |
| 1: |
| 2: for mini-batch until patience do |
| 3: // fix failures |
| Test Pass (%) | Mean Turns ( ) | Output Tokens / Episode ( ) | |||||
| Model | EvoSkill | Ours | EvoSkill | Ours | EvoSkill | Ours | |
| MATH-500 (Competition Math with Optional Python) | |||||||
| Claude Opus 5.5 | 86.4 | 87.5 | 1.00 | 1.00 | 426 | 102 | 76.0% |
| Gemini 3.8 Flash | 85.0 | 87.5 | 1.76 | 1.04 | 1,734 | 543 | 68.7% |
| Grok 4.1 Fast | 72.5 | 72.5 | 1.01 | 1.00 | 1,937 | 1,665 | 14.1% |
| Gemma 4 26B | 80.4 | 83.6 | 1.70 | 1.19 | 1,349 | 806 | 40.3% |
| Model | SKILL.md Words ( ) | Input Tok. Change | Serving Cost Change |
|---|---|---|---|
| MATH-500 (Direct math derivation fewer code retries) | |||
| Claude Opus 5.5 | 36.1% | 51.9% | |
| Gemini 3.8 Flash | 56.2% | 63.9% | |
| Grok 4.1 Fast | 8.4% | 12.1% | |
| Gemma 4 26B | 58.2% | 55.1% | |
| ALFWorld (Compact search rules fewer navigation steps) | |||
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
| Benchmark | Train / Val / Test | Action Space & Turn Cap | Deterministic Verifier ( ) |
|---|---|---|---|
| MATH-500 | 80 / 40 / 280 | Optional Python code block or final \boxed{…} (up to turns) | Exact string match or symbolic equivalence via SymPy ( s timeout). |
| ALFWorld | 79 / 39 / 280 | One TextWorld command per turn (up to turns) | Simulator goal check run after every step. |
| SpreadsheetBench | 80 / 40 / 280 | Python script ( openpyxl ) editing target workbook, or completion signal (up to turns) | Cell-by-cell cached-value match against reference workbook ( for floats; case-insensitive trimmed match for strings). |
| Benchmark / Model | Optimization Method | Batch | Val Pass | Test Pass | Test OutTok (Per-Ep) |
|---|---|---|---|---|---|
| ALFWorld | EvoSkill (Accuracy Only, ) | 8 | 89.7% | 81.4% | 26.7k (95.2) |
| (Gemma 4 26B, | Prune Directly from Seed ( ) | 8 | 71.8% ∗ | 43.2% | – (reverts to ) |
| ) | Prune Directly from Seed ( ) | 24 | 64.1% ∗ | 43.2% | – (reverts to ) |
| EfficientSkill (Start from ) | 8 | 84.6% | 81.1% | 27.8k (99.4) | |
| SpreadsheetBench | EvoSkill (Accuracy Only, ) | 8 | 50.0% | 58.2% | 3,745.1k (13,376) |
| (Gemini 3.8 Flash, | Prune Directly from Seed ( ) | 8 | 20.0% ∗ | 3.6% | – (reverts to ) |
| Package Words | Validation ( ) | Post-Peak EvoSkill | Held-Out Test ( ) | |||||
|---|---|---|---|---|---|---|---|---|
| Model | Method | SKILL.md | tools.py | Pass | OutTok | Next 6 Checkpoints | Pass | OutTok |
| Claude Opus 5.5 | Seed ( ) | 20 | 0 | 17.5% | 72.1k | – | – | – |
| EvoSkill ( ) | 1,521 | 0 | 82.5% | 71.7k | 72.5–80.0% (66k–85k) | 88.57% | 410.4k | |
| EfficientSkill | 785 | 554 | 75.0% | 35.7k | – | 88.57% | 212.9k | |
| Gemini 3.8 Flash | Seed ( ) | 20 | 0 | 2.5% | 496.5k | – | – | – |
| EvoSkill ( ) | 1,271 | 0 | 50.0% | 599.4k | 2.5–45.0% (568k–700k) | 58.21% | 3,745.1k | |
| Val (%) | EvoSkill ( ) | 10% Relative Floor | 5%-Point Slack Floor | |||||
|---|---|---|---|---|---|---|---|---|
| Benchmark | Model | Test (%) | OutTok | Test (%) | OutTok | Test (%) | OutTok | |
| MATH-500 | Claude Opus 5.5 | 86.43 | 119.2k | 87.50 | 28.6k | 87.86 | 32.4k | |
| Gemini 3.8 Flash | 85.00 | 485.5k | 87.50 | 152.1k | 86.79 | 149.7k | ||
| Grok 4.1 Fast | 72.50 | 542.3k | 72.50 | 466.1k | 73.57 | 471.4k | ||
| Gemma 4 26B | 80.36 | 377.8k | 83.57 | 225.6k | 83.57 | 205.5k | ||
| ALFWorld | Claude Opus 5.5 | 88.93 | 261.3k | 93.21 | 105.2k | 92.86 | 97.9k | |
| Benchmark | Model | (Tok/T) | Turns | Tok/T | ||
|---|---|---|---|---|---|---|
| MATH-500 | Claude Opus 5.5 | 0.0% | 76.0% | 76.0% | ||
| Gemini 3.8 Flash | 40.9% | 47.0% | 68.7% | |||
| Grok 4.1 Fast | 0.4% | 13.8% | 14.1% | |||
| Gemma 4 26B | 30.2% | 14.5% | 40.3% | |||
| ALFWorld | Claude Opus 5.5 | 28.2% | 43.9% | 59.7% | ||
| Gemini 3.8 Flash | 27.2% | 72.2% | 79.7% |
| Turns | Turn-1 | Mean Tokens / Ep | Standard Billing | Provider KV ( KV-Prov ) | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Benchmark | Model | Method | ( ) | ( ) | $/ep | $/ep | |||||
| MATH-500 | Claude Opus 5.5 | EvoSkill | 1.00 | 3,260 | 3,260 | 426 | 3,686 | $0.08085 | – | $0.08085 | – |
| EfficientSkill | 1.00 | 2,084 | 2,084 | 102 | 2,186 | $0.03892 | 51.9% | $0.03892 | 51.9% | ||
| Gemini 3.8 Flash | EvoSkill | 1.76 | 2,307 | 4,264 | 1,734 | 5,998 | $0.00084 | – | $0.00072 | – | |
| EfficientSkill | 1.04 | 1,781 | 1,867 | 543 | 2,410 | $0.00030 | 63.9% | $0.00030 | 58.4% | ||
| Grok 4.1 Fast | EvoSkill | 1.01 | 2,599 | 2,620 | 1,937 | 4,557 | $0.00149 | – | $0.00149 | – | |
| Target Spend | Batches | Est. Meta | Savings / Ep ($/ep) | Target | All-In (incl. Meta) | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Benchmark | Model | ( ) | ( ) | ( ) | Std | KV-Prov | Std | KV-Prov | Std | KV-Prov |
| MATH-500 | Claude Opus 5.5 | $46.07 | 53 | $6.36 | $0.04193 | $0.04193 | 1,099 | 1,099 | 1,251 | 1,251 |
| Gemini 3.8 Flash | $0.61 | 78 | $9.36 | $0.00054 | $0.00042 | 1,128 | 1,449 | 18,561 | 23,852 | |
| Grok 4.1 Fast | $0.78 | 28 | $3.36 | $0.00018 | $0.00018 | 4,326 | 4,363 | 22,938 | 23,138 | |
| Gemma 4 26B | $0.42 | 33 | $3.96 | $0.00085 | $0.00057 | 492 | 734 | 5,142 | 7,676 | |
| ALFWorld | Claude Opus 5.5 | $95.15 | 18 | $2.16 | $0.37101 | $0.09847 | 257 | 967 | 263 | 989 |
| EvoSkill (Unconstrained, ) | EfficientSkill (Ours, ) | |||
| FAIL : 20 turns, 36,669 out tok, 134,553 in tok | PASS : 6 turns, 11,708 out tok, 25,629 in tok | |||
| Turn | Action & Workspace Outcome | Out Tok | Action & Workspace Outcome | Out Tok |
| T1 | Prints sheet names and rows 2–5 of Sheet3 . | 595 | Loads workbook; queries defined_names.definedName AttributeError . | 327 |
| T2 | Re-prints A1:G20 coordinate tuples. | 355 | Fixes query; resolves codes2 ( D3:D14 ) and month ( E3:E14 ). | 615 |
| T3 | Inspects font, fill, and borders of empty col B . | 1,240 | Builds ’01’..’12’ map; sees naive match hit ’01’ in ’2014’ ( JAN ). | 7,480 |
| T4 | Prose only (no code executed). | 3,565 | Verifies str types across D2:E14 before writing. | 932 |
| EvoSkill (Unconstrained, ) | EfficientSkill (Ours, ) | |||
| FAIL : 1 turn, 14,649 out tok, 2,173 in tok | PASS : 2 turns, 2,597 out tok, 3,940 in tok | |||
| Turn | Action & Workspace Outcome | Out Tok | Action & Workspace Outcome | Out Tok |
| T1 | Inlines custom inspection script ( Traceback ) and narrates full plan ending in completion signal Early exit (0 saves) . | 14,649 | Calls tools.py inspection helper; prints A1:C3 and reminder that no edits are saved yet. | 226 |
| T2 | (Ended without saving at Turn 1) | Strips email suffix, writes "Sharon Smith" and "Sam Landon" to C2:C3 , saves, verifies, completes Pass . | 2,371 | |
| Total | 497 reasoning + 14,152 visible completion | 14,649 | 2,429 reasoning + 168 visible completion | 2,597 |
| EvoSkill (Unconstrained, ) | EfficientSkill (Ours, ) | |||
| PASS : 21 turns, 5,206 out tok, 49,167 in tok | PASS : 5 turns, 195 out tok, 6,479 in tok | |||
| Turn | Action & Environment Observation | Out Tok | Action & Environment Observation | Out Tok |
| T1 | go to diningtable 1 empty. | 115 | go to countertop 1 sees knife 1, mug 1, ladle 1 . | 24 |
| T2 | go to countertop 1 sees ladle 1 , walks past it . | 75 | take ladle 1 from countertop 1 ( spoon ladle alias). | 44 |
| T3 | go to sinkbasin 1 empty. | 138 | clean ladle 1 with sinkbasin 1 (in-place state change). | 41 |
| T4–T5 | Visits and opens cabinet 1 sees plate 1 . | 159 | T4: go to diningtable 1 ( table diningtable ). T5: put ladle 1 in/on diningtable 1 Goal completed! | 86 |
| EvoSkill (Unconstrained, ) | EfficientSkill (Ours, ) | |||
|---|---|---|---|---|
| PASS : 4 turns, 2,496 out tok, 9,946 in tok | PASS : 1 turn, 471 out tok, 1,775 in tok | |||
| Turn | Action & Sandbox Outcome | Out Tok | Action & Symbolic Solution | Out Tok |
| T1 | Runs SymPy sp.simplify(sp.prod(...)) returns unreduced radian product. | 570 | Pairs into , applies identity Pass . | 471 |
| T2 | Emits prose without code block no output. | 166 | (Episode finished at Turn 1) | |
| T3 | Runs Python float product ( 0.03515625 ) and checks Fraction(3,16)**2 = 9/256 . | 696 | ||
| T4 | Writes full multi-paragraph prose proof and final answer Pass . | 1,064 | ||
LLM Agent Workflow Optimization
LLM agent workflow optimization studies how to design and run multi-step agents for better task success, reliability, latency, and cost.
See all 286 papers