X-Tree: Tokenizing Reusable Experience for Efficient Agent Generalization

Sitao Cheng; Xunjian Yin; Zhiyuan Sun; Yuxuan Li; Ruiwen Zhou; Xiangru Jian; Victor Zhong

cs.AISep 26, 2026

X-Tree: Tokenizing Reusable Experience for Efficient Agent Generalization

Authors: Sitao Cheng, Xunjian Yin, Zhiyuan Sun, Yuxuan Li, Ruiwen Zhou, Xiangru Jian, Victor Zhong

Organizations: University of Waterloo · Duke University · National University of Singapore

Abstract

Multi-step agents are trained on flat action streams: SFT and RLVR weight every token uniformly and ignore the sub-procedures that recur across tasks, the hierarchy that lets humans plan top-down from reusable routines. This structure sits unused, and flat training uses each scarce trajectory less fully than its content allows. Recent agents do use that structure, but only as LLM-written skills in context, never in the weights, so their gains do not generalize beyond retrieval. We instead recover this hierarchy from the data itself and train on it, with no LLM calls. Following text tokenizers, which build a vocabulary by counting alone, we score action spans by reusability and merge canonicalized actions into a reusable eXperience tree (X-Tree). Each X-Tree node captures how a frequent and success-bearing skill is composed from sub-skills, guiding efficient generalization. We integrate X-Tree into three training settings: offline RL, with each node as a training instance; online RLVR, with an adaptive skill bonus; and on-policy self-distillation, with X-Tree as the self-teacher's privileged context. Across WebArena, ScienceWorld, and WebShop at three model scales, X-Tree improves over standard recipes at matched data and budget by up to 4.5% SR on WebArena, 5.8% SR on ScienceWorld and 4.1% success on WebShop. Matched analyses attribute the gains to the X-Tree structure and the three integrations.

Figures & tables

Figure 1: Training with experience. (a) Prior training weights every token uniformly. Harness systems use LLM-written skills in context, never in the weights. (b) We mine reusable experience by

\mathcal{X}

-Score (measuring recurrence, length, success) into the X-Tree hierarchy with zero LLM calls. Each node

v

is a recurring, success-bearing structure with depth

d_{v}

. We train X-Tree into model weights by integrating X-Tree as data (offline RL), as reward (online RLVR) and as context (OPSD). Figure 5 demonstrates a concrete X-Tree example from a trajectory.

Figure 2: Three training integrations with X-Tree. (a) Offline RL: X-Tree nodes as RL instances, partially rolled out from gold prefix and rewarded for node completion. (b) Online RLVR: an adaptive bonus for X-Tree nodes a rollout executes plus outcome signal. (c) OPSD: the same weights, given the retrieved X-Tree skills (examples in Appendix G ) as privileged context for the self-teacher.

w/o RL
Method	gitlab (187)	shopping (158)	admin (156)	reddit (112)	map (67)	wiki (14)	All (694)
Qwen2.5-7B-Instruct (base)	6.4	7.8	7.7	2.4	9.0	2.4	6.5
Go-Browse (SFT-Full w/ 2 epochs)	16.8	21.7	22.9	16.4	10.4	7.1	18.4
w/ 4 epochs (matched compute)	17.5	19.4	24.4	16.1	13.4	11.9	18.8
SFT $+$ offline RL (Ours)
Offline RL w/ Outcome reward	15.5	23.4	23.1	15.2	7.5	14.3	18.2

Table 1: Per-site SR (%) on WebArena (Qwen2.5-7B) . Go-Browse is our reproduced result. Every RL starts from SFT on half the pool. “SFT

+

offlineRL” rows keep the same offline RL stage: one rewards only the outcome of a whole trajectory, the other replaces X-Tree nodes by random spans matched in node count and length. “Plain mixing” trains on all X-Tree nodes at once without completion reward. “Curriculum” trains in order of depth. “X-Tree Full” is our recipe as designed in Eq. 2 . Our X-Tree with offline RL consistently outperforms other recipes. Bold : best per column.

Method	SR	$\Delta$
X-Tree Full	22.9	–
Tree Structure Validation
Whole trajectory (no tree)	19.9	$-3.0$
Random span	19.5	$-3.4$
Random tree	17.9	$-5.0$
Offline RL Validation

Table 2: Ablations on WebArena: Tree Structure Validation replaces X-Tree with other settings fixed; Offline RL Validation replaces the training method on identical data.

Figure 3: Online RLVR results. ScienceWorld SR on three generalization levels ( Zhang et al., 2025 ) ; WebShop success and graded score. Bold : the difference of a pair. The X-Tree integration consistently outperforms outcome-only GRPO. Statistical significance in Appendix H .

$n$	Setting	G0	G1	G2
2	outcome-only	12.9	13.7	11.8
	+ X-Tree bonus	14.6	15.5	13.0
4	outcome-only	24.2	23.9	12.1
	+ X-Tree bonus	28.0	27.5	16.5
8	outcome-only	38.6	36.2	22.8
	+ X-Tree bonus	39.8	39.5	25.6

Table 3: Rollout number

n

scaling (ScienceWorld, 1.5B) on three levels.

Warm start	outcome	$+$ X-Tree ( $\Delta$ )
500 Samples	73.8	74.9 $(+1.1)$
1,016 Success	74.9	78.1 $(+3.2)$
1,824 All	76.4	77.4 $(+1.0)$

Table 4: Resource-matched runs (WebShop, 1.5B). All rows use same resources and differ only in reward.

# Traj	Skills	Success (%)
50	9	68.7
100	11	75.2
200	16	72.9
500	24	74.4
1,824	48	76.5

Table 5: Mining-corpus size (WebShop, 1.5B). X-Tree is mined from

N

random trajectories of the corpus.

Figure 4: Validating OPSD with X-Tree. ScienceWorld SR on three levels ( Zhang et al., 2025 ) ; WebShop success and graded score. OPSD with X-Tree changes only the privileged context of the self-teacher. X-Tree is comparable with LLM-written bank. Statistical significance in Appendix H .

Appendix figures & tables24 assets

Supplementary material from the paper’s appendix.

Appendix

X-Tree Mining (Eq. 1 )
Setting	WebArena	ScienceWorld	WebShop
$p_{\ell}$ , $p_{s}$	1, 1	1, 1	1, 1
Success floor $\epsilon$	0.001	0.001	0.001
$\eta$	0.3	0.5	0.5
Skill cap	256	80	60
Max. skill length (actions)	12	8	8

Appendix

Table 6: X-Tree construction and SFT hyperparameters. The canonicalization and merging are per environment. Merging stops at the compression constraint § 3.1 or the skill cap.

Setting	WebArena (offline)	ScienceWorld	WebShop
Rollouts per group $n$	2	8	8
Optimizer steps	805	100	150
Units per step	6 nodes	8 tasks	16 tasks
Learning rate	$5{\times}10^{-7}$	$1{\times}10^{-6}$	$1{\times}10^{-6}$
KL to the warm start	none	0.01	0.01
Mini-batch	2	256	64

Appendix

Table 7: RL hyperparameters for the three environments.

Setting	MiniWoB++
Backbones	Qwen2.5-7B, Qwen3-8B, Qwen3-14B
Mining corpus	episodes of the evaluated backbone on the 53 training task families, 40 per family
Trajectories (successful)	2,125 (886) / 2,120 (1,116) / 2,120 (1,239)
Filtering	none, failures kept with their label
Canonical action	verb $\langle$ role $\rangle$ , role from the DOM (Table 10 )
$p_{\ell}$ , $p_{s}$	1, 1.5

Appendix

Table 8: X-Tree construction and deployment for the prompted-agents setting (Table 15 ). Merging follows Eq. 1 and stops at the compression constraint or the skill cap.

Raw action	Canonical	Kept in slots
Verb , from the BrowserGym function name
click(bid) , dblclick(bid)	click	element id
fill(bid, text) , type(bid, text)	type	element id, typed text
select_option(bid, opt)	select	element id, option value
press(bid, key) , keyboard_press(key)	key	element id, key name
hover , focus , clear , scroll	unchanged	element id

Appendix

Table 9: Canonicalization on WebArena. A BrowserGym action string to verb

\langle

role

\rangle

: verb is from the function name, role from the accessibility-tree role of the element the action addresses.

Raw action or element	Canonical	Kept in slots
Verb
click(ref) , submit(ref) , clear(ref)	click , submit , clear	element label
noop	noop	—
type(ref, text)	type	typed text, element label
select(ref, value)	select	option value, element label
key(k)	key	key name

Appendix

Table 10: Canonicalization on MiniWoB++. The verb is from the agent’s primitive action. The role is from the addressed DOM element by the ordered rule below, first matching row wins. The element label (nearest label, ARIA label, text, placeholder, value or id) is kept in the slots. In our corpus the environment reports every input as a typed tag, so the last rule assigns its role.

Command	Canonical primitive	Kept in slots
ScienceWorld
focus on $X$	focus $\langle$ $c(X)$ $\rangle$	$X$
go to $X$ , teleport to $X$	navigate $\langle$ $X$ $\rangle$ if $X$ is a room, else navigate $\langle$ $c(X)$ $\rangle$	$X$ , teleport flag
open $X$ , close $X$	open $\langle$ $c(X)$ $\rangle$ , close $\langle$ $c(X)$ $\rangle$	$X$
pick up $X$	take $\langle$ $c(X)$ $\rangle$	$X$
put down $X$ , drop $X$	put $\langle$ ground $\rangle$	$X$ , surface form

Appendix

Table 11: Canonicalization on ScienceWorld and WebShop. Both environments take text commands, so the rule is a template match rather than a role lookup.

X

and

Y

are object phrases, and

c(\cdot)

is the nine-class lexicon defined in the text.

Figure 5: An example trajectory “tokenized” by the X-Tree (ScienceWorld). Task: measure the temperature of unknown substance B in the bedroom, then place it in the green box above 50 degrees and in the blue box below. Rows are the trajectory’s 18 actions with their canonical tokens. The tree beside them is the X-Tree over the actions, with depth growing to the right, and each shaded band is one token of the tiling. S2 (purple) recurs inside two different tokens.

Canonical node	Depth	Occ.	Grounded occurrence	Reading
type $\langle$ text_input $\rangle$ $\to$ type $\langle$ text_input $\rangle$ $\to$ click $\langle$ button $\rangle$	2	209	fill “From ” 01/01/2025; fill “To ” 03/24/2025; click “Show Report” (shop admin); fill “From” Times Square; fill “To” Central Park; click “Go” (map)	complete a two-field form and submit: a report filter and a route query are the same node
type $\langle$ text_input $\rangle$ $\to$ click $\langle$ button $\rangle$	1	430	fill searchbox “Search for projects, issues, etc.” project ; click “Search” (GitLab)	type a query and submit it
type $\langle$ text_input $\rangle$ $\to$ type $\langle$ text_input $\rangle$	1	88	fill “From” Grand Central Station; fill “To” Empire State Building (map)	the two-field entry that recurs inside the first row
click $\langle$ link $\rangle$ $\to$ click $\langle$ link $\rangle$ $\to$ click $\langle$ link $\rangle$	2	161	click “Forums”; click a page link; click the next page (Reddit)	navigate, then paginate
click $\langle$ link $\rangle$ $\to$ click $\langle$ link $\rangle$	1	1,219	click “SALES”; click a submenu entry (shop admin)	menu navigation

Appendix

Table 12: Case study (a): X-Tree nodes in WebArena. Canonical nodes are written in the verb

\langle

role

\rangle

tokens of Table 9 . Occ. counts the occurrence of the node. The grounded occurrence reads element labels from the page’s accessibility tree.

Canonical node	Depth	Occ.	Grounded occurrence	Reading
S8 $=$ S2 $+$ navigate $\langle$ kitchen $\rangle$	3	415	open door to hallway; go to hallway; open door to kitchen; go to kitchen	go to the kitchen via the hallway
S31 $=$ put $\langle$ device $\rangle$ $+$ S30	4	151	move metal pot containing gallium to stove; activate stove; examine gallium; use thermometer on gallium; examine; use thermometer	heat a substance and monitor its temperature
S60 $=$ navigate $\langle$ foundry $\rangle$ $+$ S59	6	25	go to foundry; open blast furnace; move metal pot to blast furnace; activate blast furnace; examine lead; use thermometer on lead; examine; use thermometer	melt a metal and monitor it
S5 $=$ observe $\langle$ room $\rangle$ $+$ S4	4	381	look around; connect battery anode to black wire terminal 1; connect battery cathode to blue wire terminal 1; connect black wire terminal 2 to cathode in buzzer; …	wire a circuit from battery through wires to a load
S27 $=$ S26 $+$ take $\langle$ animal $\rangle$	4	241	open door to outside; go to outside; look around; focus on baby wolf; pick up baby wolf	find and collect an animal outdoors
S43 $=$ S7 $+$ put $\langle$ container $\rangle$	4	90	open door to hallway; go to hallway; open door to workshop; go to workshop; move baby wolf in inventory to blue box	carry the animal to a container in another room

Appendix

Table 13: Case study (b): X-Tree nodes in ScienceWorld. ScienceWorld nodes are long, so each canonical node is written as its merge of two children, in the primitives of Table 11 . Occ. counts occurrences in the

1{,}673

AgentTraj-L trajectories.

Canonical node	Depth	Occ.	Success	Reading
search $\langle$ query $\rangle$ $\to$ click $\langle$ item $\rangle$ $\to$ click $\langle$ option-color $\rangle$ $\to$ click $\langle$ buy $\rangle$	3	94	0.81	engage the goal’s color option before buying
search $\langle$ query $\rangle$ $\to$ click $\langle$ item $\rangle$ $\to$ click $\langle$ option-size $\rangle$ $\to$ click $\langle$ option-color $\rangle$ $\to$ click $\langle$ buy $\rangle$	4	25	0.92	two typed options, both named by the goal
search $\langle$ query $\rangle$ $\to$ click $\langle$ item $\rangle$ $\to$ click $\langle$ option-flavor-name $\rangle$	2	31	0.97	a typed option the goal names
click $\langle$ option-color $\rangle$ $\to$ click $\langle$ buy $\rangle$	1	139	0.84	the closing move once the right option is set
search $\langle$ query $\rangle$ $\to$ click $\langle$ item $\rangle$ $\to$ click $\langle$ buy $\rangle$	2	741	0.57	buy without touching options: a universal macro at the corpus mean
search $\langle$ query $\rangle$ $\to$ click $\langle$ item $\rangle$	1	1,770	0.57	a universal macro

Appendix

Table 14: Case study (c): X-Tree nodes in WebShop. Canonical nodes are written in the primitives of Table 11 . Occ. counts occurrences in the

1{,}824

ETO trajectories. ETO is the corpus with failures, so the fraction of a node’s occurrences in successful trajectories (corpus mean

0.56

) is reported.

Skill bank	Qwen2.5-7B	Qwen3-8B	Qwen3-14B
no skills	44.9	60.4	68.1
AWM	41.4	59.9	69.0
SkillWeaver	47.6	61.3	68.3
WALT	46.3	58.1	68.5
WebXSkill	43.8	59.8	68.6
X-Tree	49.8	65.1	72.6

Appendix

Table 15: X-Tree as advice in context (MiniWoB++, SR, %). Only the skill bank in context changes. Four LLM-written baselines and X-Tree.

SFT row	1.5B	3B	7B
one standard action	11.7	16.7	20.8
length-matched span	12.0	17.4	19.3
an X-Tree node	16.6	19.1	29.9

Appendix

Table 16: SFT at 1.5B/3B/7B on ScienceWorld (SR pooled over G0 and G1).

Method	official SR	SR (main text)
base model	5.8	6.5
SFT-Full (Go-Browse), 2 epochs	17.2	18.4
SFT-Full, 4 epochs	17.4	18.8
whole-trajectory GRPO	18.7	19.9
whole-trajectory GRPO, binary outcome reward	17.6	18.2
random-node GRPO	18.5	19.5

Appendix

Table 17: Official evaluator and normalized SR (WebArena, Qwen2.5-7B, mean over three passes).

Backbone	SFT-Full	$+$ X-Tree recipe	$\Delta$
Qwen2.5-7B	18.4	22.9	$+4.5$
Qwen2.5-14B	24.1	25.7	$+1.6$
GLM-4-9B	14.9	16.9	$+2.0$

Appendix

Table 18: Scale and model-family transfer of offline RL with X-Tree (WebArena, SR, %).

Figure 6: What the trained policies execute at test time (1.5B, G0 fold). Left: occurrences of executed X-Tree skills by depth. Right: properties of the trained models.

Figure 7: How

\lambda_{g}

anneals during training (ScienceWorld). Per training step, the group-mean weight

\lambda_{g}

of Eq. 3 (solid) and the training win rate (dashed).

\lambda_{0}{=}0.75

and

w_{\mathrm{ref}}{=}0.4

are dotted. 3B: the three runs behind Figure 3 (faint) and their mean (bold). 1.5B: one run.

Setting	$\lambda_{0}$	$w_{\mathrm{ref}}$	G0 / G1
outcome-only GRPO	–	–	38.6 / 36.2
gentler weight	0.5	0.4	39.4 / 37.6
front-loaded, anneals sooner	1.5	0.25	38.9 / 36.4
selected (Ours)	0.75	0.4	39.8 / 39.5

Appendix

Table 19: Sensitivity to

\lambda_{0}

and

w_{\mathrm{ref}}

(ScienceWorld, 1.5B,

n{=}8

, SR on G0 / G1).

Skills	# Match	G0	G1	G2
Empty (outcome-only)	–	38.6	36.2	22.8
Random spans	576	38.4 $(-0.2)$	38.4 $(+2.2)$	–
Random tree	2,109	38.1 $(-0.5)$	36.8 $(+0.6)$	26.7 $(+3.9)$
X-Tree	2,698	39.8 $(+1.2)$	39.5 $(+3.3)$	25.6 $(+2.8)$

Appendix

Table 20: Ablation of X-Tree (ScienceWorld, 1.5B). SR (%), difference to outcome-only in parentheses. “#Match” is the number of nodes the bonus can match.

	full corpus	testing task types removed
trajectories after the length filter	1,673	1,248
# X-Tree nodes	80	80 (66 kept, 14 new)
grounded skills the bonus reward can match	2,698	1,305
G2 SR, outcome $+$ X-Tree bonus (three runs)	25.6	24.8
G2 SR, outcome-only GRPO	22.8

Appendix

Table 21: Results with held-out tasks removed from the mining corpus (ScienceWorld, 1.5B). Top: X-Tree changes from the held-out tasks removed. Bottom: performance with X-Tree original and re-mined, everything else fixed.

### TASK: goal requires a color option ###

The instruction names a color. Expert episodes that selected the goal’s color on the product page before buying, and how they fared:

- search<query> -> click<item> -> click<option-color> -> click<buy>: 81% success (94 episodes, ABOVE base) – search for the product; open a result item; select the goal’s color option; click Buy Now.

- click<option-color> -> click<option> -> click<buy>: 62% success (34 episodes, ABOVE base) – select the goal’s color option; select an option the goal does NOT ask for (wasted step); click Buy Now.

[four more chains]

Plan: search with the product words plus the color value; open a matching item; click the exact color value the instruction asks for (it must be visible on the page); select any other required options the same way; then Buy Now.

Appendix

Table 22: The X-Tree rendering example for the color option , as the self-teacher reads it.

Teacher’s bank	Success
outcome-only RLVR: $74.0$
Empty (no privileged text)	74.2
X-Tree, words scrambled	76.5
Mined X-Tree	77.1

Appendix

Table 23: Ablation on skill bank to the self-teacher (WebShop, 3B).

Scale	Setting	G0	G1	held-out G2
1.5B	Outcome-only GRPO	$38.6\pm 2.5$	$36.2\pm 0.3$	$22.8\pm 1.3$
	+ X-Tree bonus	$39.8\pm 1.0$	$39.5\pm 1.7$	$25.6\pm 1.3$
	+ OPSD, LLM-written bank	$39.4\pm 1.7$	$38.4\pm 2.1$	$25.4\pm 2.0$
	+ OPSD, X-Tree bank	$38.7\pm 2.7$	$35.9\pm 1.8$	$26.3\pm 1.8$
3B	Outcome-only GRPO	$54.6\pm 6.0$	$48.8\pm 3.1$	$24.0\pm 3.8$
	+ X-Tree bonus	$56.4\pm 3.7$	$52.6\pm 5.6$	$26.2\pm 2.3$

Appendix

Table 24: Spread behind ScienceWorld. SR (%) as mean

\pm

sample standard deviation behind Figure 3 and the OPSD results of Figure 4 (top) and Table 3 (bottom, 1.5B).

Scale	Setting	Success	Graded	$\Delta$ ( $t$ )
1.5B	RLVR, outcome-only	$73.8\pm 1.9$	$88.0\pm 1.0$	—
	$+$ X-Tree bonus	$76.5\pm 1.6$	$89.6\pm 1.1$	$+2.7$ (7.0)
	OPSD, LLM-written bank	$74.7\pm 1.9$	$87.7\pm 1.0$	$+0.9$ (0.8)
	OPSD, X-Tree bank	$73.8\pm 1.7$	$87.4\pm 1.1$	$+0.0$ (0.1)
3B	RLVR, outcome-only	$74.0\pm 3.0$	$87.7\pm 1.6$	—
	$+$ X-Tree bonus	$75.9\pm 3.5$	$88.4\pm 1.6$	$+1.9$ (2.4)

Appendix

Table 25: Spread behind WebShop. Success and graded score (%) as mean

\pm

sample standard deviation behind Figure 3 and Figure 4 .

\Delta

(

t

) is the paired success difference against the outcome-only row of the same scale and its

t

-statistic.

	Anchor	outcome-only	$+$ X-Tree
X-Tree mined from the warm start trajectories
500 samples	24.2	$73.8\pm 1.9$	$74.9\pm 1.0$
1,016 successful	33.1	$74.9\pm 1.5$	$78.1\pm 2.1$
1,824 all	32.5	$76.4\pm 4.1$	$77.4\pm 1.1$
$N$ trajectories in mining corpus. All trained from the $500$ -sample warm start
$N{=}50$ , 9 skills	-	-	$68.7\pm 0.7$

Appendix

Table 26: Resource-matched controls and mining-corpus size, with their spreads (WebShop, 1.5B). Success (%) as mean

\pm

sd over three runs. Top : both runs of a row share the warm start and the RL data, and X-Tree is mined from the SFT trajectory. “Anchor” is the warm start’s own success before RL. Bottom : the X-Tree is mined from

N

random trajectories.

LLM Agent Skill Learning

Large language model agent skill learning studies how agents acquire, represent, reuse, and refine procedures for multi-step tasks.

See all 402 papers

LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Oct 23, 2026Shen Liu, Ze Zhou, Zhenguo Xu +3LLM Agent Skill Retrieval LLM Grounding
North China Institute of Computer System Engineering, Beijing, China · Beijing University of Aeronautics and Astronautics, Beijing, China · University of Science and Technology of China, Hefei, China +1
SkillX: Automatically Constructing Skill Knowledge Bases for Agents
Apr 6, 2026Chenxi Wang, Zhuoyun Yu, Xin Xie +8LLM Agent Skill Learning LLM Agents
1Zhejiang University · 2Ant Digital Technologies, Ant Group.
Branch2Skill: Efficient Skill Evolution Through Reasoning Trees
Aug 9, 2026Yanwei Ren, Haotian Zhang, Likang Xiao +5LLM Agent Skill Learning Agentic Reasoning
School of Artificial Intelligence, Beihang University, Beijing, China · Hangzhou International Innovation Institute, Beihang University, Hangzhou, China · The Hong Kong Polytechnic University, Hong Kong SAR, China +3

Oct 23, 2026J/K move · Enter open · S save

LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents

Shen Liu, Ze Zhou, Zhenguo Xu, Shaopu Wang+2

North China Institute of Computer System Engineering, Beijing, China · Beijing University of Aeronautics and Astronautics, Beijing, China · University of Science and Technology of China, Hefei, China+1

LLM Agent Skill Retrieval LLM Grounding Long-Horizon LLM Agents

Open paper AskPDF arXiv

Recursive large language model (LLM) agents solve long-horizon tasks through decomposition, execution, and environmental feedback. Yet after producing an atomic subtask, they often invoke the same large model again to translate a familiar instruction into an executable action. We observe that high-level planning is task-specific, whereas leaf-level action grounding recurs across tasks and can be checked against environment-admitted actions. We introduce LeafSkill, an external reuse layer that retrieves semantically related grounding experience, binds action templates to current objects and admissible actions, and uses a lightweight 1.7B selector to choose a bound action or defer the unchanged subtask to the original agent. Skills remain outside the model parameters, allowing a frozen selector to use post-training additions. Under lexical and syntactic paraphrases, exact matching retrieves none of the reference actions, while semantic retrieval preserves 96.0% and 97.0% candidate coverage; the frozen selector adopts 97.34% of unseen skill types. On an independent confirmation set of 200 Robotouille tasks, LeafSkill completes 137 tasks versus 121 for ReCAP (68.5% vs. 60.5%, exact McNemar p=.0259). On the 106 tasks completed by both, it reduces Qwen3.5-27B tokens by 14.42% and calls by 14.08%. A separate held-out repeat yields a consistent 9.0-point success gain and 18.98% token reduction. Reusable leaf grounding can thus improve task completion while reducing host-model inference without replacing high-level planning.

Abstract

LLM Agent Skill Learning

Explore similar work

LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents

SkillX: Automatically Constructing Skill Knowledge Bases for Agents

Branch2Skill: Efficient Skill Evolution Through Reasoning Trees