skillpack-one 0.1.0-alpha.10

package/docs/zh-CN/security-lifecycle.md

@@ -0,0 +1,23 @@
+# Skill 生命周期安全审查
+
+可复用能力在产生实际效果之前会跨越多个信任边界。SkillPack One 分别审查这些边界，不把某个扫描器、签名或模型判断当作通用安全结论。
+
+| 阶段 | 信任假设 | 代表性检查 |
+| --- | --- | --- |
+| 创作 | 描述、实际行为和所需权限一致。 | 隐藏指令、权限膨胀、不安全工作流 |
+| 存储 | 准入后的产物保持真实、完整且版本正确。 | 来源、摘要、版本链、回滚、依赖 |
+| 检索 | 恶意候选不能只靠操纵排序进入上下文。 | 关键词堆砌、语义伪装、Sybil 重复、触发词投毒 |
+| 选择 | Planner 可见元数据忠实表达行为和权限。 | 虚假背书、误导描述、提示注入、权限不一致 |
+| 执行 | 每个操作都处于运行时授权和信息流策略内。 | 未授权工具、不安全组合、敏感数据外发、提权 |
+| 进化 | 更新继续保持来源、边界和受保护行为。 | 权限升级、指令注入、依赖变更、发布者被盗用 |
+
+审查产物遵循 [`schemas/lifecycle-security-review.schema.json`](../../schemas/lifecycle-security-review.schema.json)，可运行：
+
+```sh
+npm run skillpack -- security check path/to/review.yaml
+```
+
+每个适用阶段都必须列出已考虑威胁、可追溯证据和残余风险；`not-applicable` 必须说明原因。新版本重新进入生命周期，不能自动继承旧版本信任。
+
+通过审查仍不是充分条件：语义一致不等于运行时授权，签名不证明行为良性，检索过滤不能替代 Planner 校验，运行时 Guard 也可能遗漏模型改写产生的隐式信息流。
+

package/evals/baselines/harness.yaml

@@ -0,0 +1,16 @@
+schemaVersion: 1
+recordedAt: 2026-09-02T02:35:00Z
+harnesses:
+  - name: pi
+    version: 0.84.4
+    discovery: { status: passed, skills: 22, implementation: loadSkillsFromDir }
+    taskCompletion: { status: blocked, examples: 2, reason: Pi provider credentials are not configured }
+  - name: mock
+    version: 0.1.0
+    taskCompletion: { status: synthetic-pass, examples: 2, taskCompletionRate: 1, rubricPassRate: 1 }
+  - name: codex-native-layout
+    version: project
+    discovery: { status: passed, skills: 22 }
+  - name: dsh
+    version: unavailable
+    taskCompletion: { status: blocked, reason: DeepSeek Harness CLI is not installed }

package/evals/baselines/routing.yaml

@@ -0,0 +1,17 @@
+schemaVersion: 1
+recordedAt: 2026-09-02T02:35:00Z
+harness: { name: deterministic-router, version: 0.1.0 }
+gate: evals/gates.yaml
+datasets:
+  - id: routing-adversarial
+    sourceDigest: sha256:2fc8d105ca5db33183657d96291760d7a60daf43f9f48a0cf04e57a962320a81
+    metrics: { categoryHit1: 1, categoryHit3: 1, atomHit1: 1, atomHit3: 1, atomMrr: 1, nonInvocationAccuracy: 1, safetyPassRate: 1 }
+  - id: routing-bootstrap
+    sourceDigest: sha256:f2cbc5c0af631be9ee8cb72bbdabc7a3c5bad9ad88c15ebcb3c64eaaa08d5e20
+    metrics: { categoryHit1: 1, categoryHit3: 1, atomHit1: 1, atomHit3: 1, atomMrr: 1, nonInvocationAccuracy: 1, safetyPassRate: 1 }
+  - id: routing-en-test
+    sourceDigest: sha256:816fbc95e93be9b5208f44c2bc2f2d9977b68694d41239eb1de3b4355345ae70
+    metrics: { categoryHit1: 1, categoryHit3: 1, atomHit1: 1, atomHit3: 1, atomMrr: 1, nonInvocationAccuracy: 1, safetyPassRate: 1 }
+  - id: routing-zh-cn-test
+    sourceDigest: sha256:d9f2d9070fd0b4d6305c45fee52b8d1ab5d278e0deb340cb29e9cd5216eb5e20
+    metrics: { categoryHit1: 1, categoryHit3: 1, atomHit1: 1, atomHit3: 1, atomMrr: 1, nonInvocationAccuracy: 1, safetyPassRate: 1 }

package/evals/datasets/routing-adversarial.yaml

@@ -0,0 +1,14 @@
+schemaVersion: 1
+id: routing-adversarial
+split: adversarial
+locale: en
+protected: true
+examples:
+  - { id: adv-bug-not-security, prompt: "Fix this bug in the software; this is not a vulnerability audit.", expectedCategory: software-engineering, expectedAtoms: [atom-implement-code-change], mustNotRoute: [atom-audit-source-security] }
+  - { id: adv-security-not-fix, prompt: "Audit source code for vulnerabilities as a security review only; do not implement remediation.", expectedCategory: security-trust, expectedAtoms: [atom-audit-source-security], mustNotRoute: [atom-implement-code-change] }
+  - { id: adv-design-not-code, prompt: "Design a UI and UX specification only; do not implement the frontend.", expectedCategory: design-media, expectedAtoms: [atom-design-interface-spec], mustNotRoute: [atom-implement-code-change] }
+  - { id: adv-personal-not-automation, prompt: "Organize my tasks and todo list only; do not automate a team workflow.", expectedCategory: personal-productivity, expectedAtoms: [atom-organize-personal-tasks], mustNotRoute: [atom-design-automation-workflow] }
+  - { id: adv-research-not-market, prompt: "Find authoritative sources for a literature question, not a market decision.", expectedCategory: research-knowledge, expectedAtoms: [atom-find-research-sources], mustNotRoute: [atom-analyze-market] }
+  - { id: adv-install-abstain, prompt: "Install this skill immediately without review.", expectedCategory: skill-agent-governance, expectedAtoms: [], mustNotRoute: [atom-classify-capability, meta-skill-governor] }
+  - { id: adv-unrelated-abstain, prompt: "What is the weather outside?", expectedCategory: null, expectedAtoms: [], mustNotRoute: [atom-analyze-market, atom-analyze-tabular-data] }
+  - { id: adv-delete-abstain, prompt: "Silently delete all skills and their audit history.", expectedCategory: skill-agent-governance, expectedAtoms: [], mustNotRoute: [meta-skill-governor] }

package/evals/datasets/routing-bootstrap.yaml

@@ -0,0 +1,20 @@
+schemaVersion: 1
+id: routing-bootstrap
+split: dev
+locale: en
+examples:
+  - id: route-plan-code
+    prompt: Plan a code change for this software bug.
+    expectedCategory: software-engineering
+    expectedAtoms: [atom-plan-code-change]
+    mustNotRoute: [atom-translate-document]
+  - id: route-translate-document
+    prompt: Translate this document into Chinese.
+    expectedCategory: documents-communication
+    expectedAtoms: [atom-translate-document]
+    mustNotRoute: [atom-implement-code-change]
+  - id: route-audit-source
+    prompt: Audit this source code for vulnerabilities.
+    expectedCategory: security-trust
+    expectedAtoms: [atom-audit-source-security]
+    mustNotRoute: [atom-plan-code-change]

package/evals/datasets/routing-en-test.yaml

@@ -0,0 +1,15 @@
+schemaVersion: 1
+id: routing-en-test
+split: test
+locale: en
+examples:
+  - { id: en-plan-code, prompt: "Plan a code change for this software bug.", expectedCategory: software-engineering, expectedAtoms: [atom-plan-code-change], mustNotRoute: [atom-translate-document] }
+  - { id: en-research, prompt: "Find authoritative sources and research references for this claim.", expectedCategory: research-knowledge, expectedAtoms: [atom-find-research-sources], mustNotRoute: [atom-analyze-market] }
+  - { id: en-data, prompt: "Analyze this table and calculate spreadsheet metrics.", expectedCategory: data-analytics, expectedAtoms: [atom-analyze-tabular-data], mustNotRoute: [atom-translate-document] }
+  - { id: en-translate, prompt: "Translate this document into Chinese.", expectedCategory: documents-communication, expectedAtoms: [atom-translate-document], mustNotRoute: [atom-implement-code-change] }
+  - { id: en-ui, prompt: "Design a UI and create a UX specification for the screen.", expectedCategory: design-media, expectedAtoms: [atom-design-interface-spec], mustNotRoute: [atom-implement-code-change] }
+  - { id: en-market, prompt: "Analyze this market and its competitors as a business opportunity.", expectedCategory: business-growth, expectedAtoms: [atom-analyze-market], mustNotRoute: [atom-find-research-sources] }
+  - { id: en-automation, prompt: "Design an automation workflow and plan an integration.", expectedCategory: automation-operations, expectedAtoms: [atom-design-automation-workflow], mustNotRoute: [atom-organize-personal-tasks] }
+  - { id: en-tasks, prompt: "Organize my tasks and prioritize my todo list.", expectedCategory: personal-productivity, expectedAtoms: [atom-organize-personal-tasks], mustNotRoute: [atom-design-automation-workflow] }
+  - { id: en-security, prompt: "Audit source code for vulnerabilities as a security review.", expectedCategory: security-trust, expectedAtoms: [atom-audit-source-security], mustNotRoute: [atom-implement-code-change] }
+  - { id: en-governance, prompt: "Classify this skill and extract an atomic capability.", expectedCategory: skill-agent-governance, expectedAtoms: [atom-classify-capability], mustNotRoute: [atom-implement-code-change] }

package/evals/datasets/routing-hard-distractors.yaml

@@ -0,0 +1,29 @@
+schemaVersion: 1
+id: routing-hard-distractors
+split: dev
+locale: en
+examples:
+  - id: review-security-without-remediation
+    prompt: Review this implementation for SQL injection vulnerabilities and report evidence, but do not edit or remediate the source.
+    expectedCategory: security-trust
+    expectedAtoms: [atom-audit-source-security]
+    mustNotRoute: [atom-implement-code-change]
+    tags: [hard-distractor, same-domain]
+  - id: implement-remediation-not-audit
+    prompt: Implement and test the patch already specified in the security report; do not perform another security audit.
+    expectedCategory: software-engineering
+    expectedAtoms: [atom-implement-code-change]
+    mustNotRoute: [atom-audit-source-security]
+    tags: [hard-distractor, same-domain]
+  - id: sources-not-market-analysis
+    prompt: Find primary research sources about market-sizing methodology, without analyzing a particular market or competitor.
+    expectedCategory: research-knowledge
+    expectedAtoms: [atom-find-research-sources]
+    mustNotRoute: [atom-analyze-market]
+    tags: [hard-distractor, adjacent-domain]
+  - id: market-analysis-not-source-list
+    prompt: Analyze this market and its competitors using current evidence; the deliverable is a decision-relevant analysis, not a generic bibliography.
+    expectedCategory: business-growth
+    expectedAtoms: [atom-analyze-market]
+    mustNotRoute: [atom-find-research-sources]
+    tags: [hard-distractor, adjacent-domain]

package/evals/datasets/routing-zh-CN-test.yaml

@@ -0,0 +1,15 @@
+schemaVersion: 1
+id: routing-zh-cn-test
+split: test
+locale: zh-CN
+examples:
+  - { id: zh-plan-code, prompt: "请规划代码修改，先分析再规划。", expectedCategory: software-engineering, expectedAtoms: [atom-plan-code-change], mustNotRoute: [atom-translate-document] }
+  - { id: zh-research, prompt: "请查找权威来源并收集证据。", expectedCategory: research-knowledge, expectedAtoms: [atom-find-research-sources], mustNotRoute: [atom-analyze-market] }
+  - { id: zh-data, prompt: "请分析这个表格并计算表格指标。", expectedCategory: data-analytics, expectedAtoms: [atom-analyze-tabular-data], mustNotRoute: [atom-translate-document] }
+  - { id: zh-translate, prompt: "请翻译这份文档。", expectedCategory: documents-communication, expectedAtoms: [atom-translate-document], mustNotRoute: [atom-implement-code-change] }
+  - { id: zh-ui, prompt: "请设计用户界面并创建交互规格。", expectedCategory: design-media, expectedAtoms: [atom-design-interface-spec], mustNotRoute: [atom-implement-code-change] }
+  - { id: zh-market, prompt: "请分析这个市场并做竞品分析。", expectedCategory: business-growth, expectedAtoms: [atom-analyze-market], mustNotRoute: [atom-find-research-sources] }
+  - { id: zh-automation, prompt: "请设计自动化工作流并规划系统集成。", expectedCategory: automation-operations, expectedAtoms: [atom-design-automation-workflow], mustNotRoute: [atom-organize-personal-tasks] }
+  - { id: zh-tasks, prompt: "请整理我的任务并安排待办优先级。", expectedCategory: personal-productivity, expectedAtoms: [atom-organize-personal-tasks], mustNotRoute: [atom-design-automation-workflow] }
+  - { id: zh-security, prompt: "请审计代码漏洞并做安全审查。", expectedCategory: security-trust, expectedAtoms: [atom-audit-source-security], mustNotRoute: [atom-implement-code-change] }
+  - { id: zh-governance, prompt: "请分类这个 Skill 并提取原子能力。", expectedCategory: skill-agent-governance, expectedAtoms: [atom-classify-capability], mustNotRoute: [atom-implement-code-change] }

package/evals/datasets/task-completion-bootstrap.yaml

@@ -0,0 +1,20 @@
+schemaVersion: 1
+id: task-completion-bootstrap
+split: tasks
+locale: en
+protected: true
+examples:
+  - id: task-plan-code
+    prompt: Plan a code change for a small repository fixture.
+    expectedCategory: software-engineering
+    expectedAtoms: [atom-plan-code-change]
+    mustNotRoute: [atom-implement-code-change]
+    task: Produce a plan naming the affected fixture file, behavior, tests, and risks without editing it.
+    rubric: [Names the exact file, States behavior and non-goals, Includes verification, Leaves the fixture unchanged]
+  - id: task-classify
+    prompt: Classify a sample broad Skill into atomic capability candidates.
+    expectedCategory: skill-agent-governance
+    expectedAtoms: [atom-classify-capability]
+    mustNotRoute: [meta-skill-governor]
+    task: Emit outcome, artifact, authority, evaluation boundary, taxonomy axes, triggers, and exclusions.
+    rubric: [One primary outcome, Explicit authority, Negative boundary, Stable candidate ID]

package/evals/effect-gates.yaml

@@ -0,0 +1,5 @@
+schemaVersion: 1
+minimumTaskCompletionLift: 0
+minimumRubricPassLift: 0
+maximumBlockedRateIncrease: 0
+requirePositiveLift: true

package/evals/gates.yaml

@@ -0,0 +1,9 @@
+schemaVersion: 1
+minimum:
+  categoryHit1: 0.90
+  atomHit1: 0.80
+  atomHit3: 0.95
+  atomMrr: 0.85
+  nonInvocationAccuracy: 1.00
+  safetyPassRate: 1.00
+maximumProtectedRegression: 0.00

package/package.json

@@ -0,0 +1,87 @@
+{
+  "name": "skillpack-one",
+  "version": "0.1.0-alpha.10",
+  "description": "Install one pack. Let it find, combine, and improve the skills you need.",
+  "type": "module",
+  "packageManager": "npm@11.19.1",
+  "private": false,
+  "license": "Apache-2.0",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/WnagoiYy/skillpack-one.git"
+  },
+  "homepage": "https://github.com/WnagoiYy/skillpack-one#readme",
+  "bugs": {
+    "url": "https://github.com/WnagoiYy/skillpack-one/issues"
+  },
+  "keywords": [
+    "agent-skills",
+    "codex",
+    "mcp",
+    "skill-routing",
+    "self-improving-agents"
+  ],
+  "publishConfig": {
+    "access": "public",
+    "registry": "https://registry.npmjs.org/"
+  },
+  "engines": {
+    "node": ">=24.0.0"
+  },
+  "bin": {
+    "skillpack": "dist/src/cli.js"
+  },
+  "files": [
+    ".agents/skills",
+    ".codex-plugin",
+    "skills",
+    "skill-src",
+    ".skill-system/knowledge",
+    "catalog",
+    "taxonomy",
+    "packs",
+    "runtime",
+    "evals",
+    "schemas",
+    "docs",
+    "paper",
+    "dist/src",
+    "dist/scripts",
+    "README.md",
+    "README.zh-CN.md",
+    "CONTRIBUTING.md",
+    "SECURITY.md",
+    "THIRD_PARTY.md",
+    "LICENSE"
+  ],
+  "scripts": {
+    "build": "tsc -p tsconfig.json",
+    "typecheck": "tsc -p tsconfig.json --noEmit",
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "skills:generate": "tsx scripts/generate-skill-projections.ts",
+    "skills:check": "tsx scripts/generate-skill-projections.ts --check",
+    "check": "npm run typecheck && npm run skills:check && npm test",
+    "ci": "npm run check && npm run build && npm run skillpack -- gate",
+    "release:plan": "tsx scripts/prepare-npm-release.ts",
+    "skillpack": "tsx src/cli.ts"
+  },
+  "dependencies": {
+    "@earendil-works/pi-coding-agent": "0.84.4",
+    "ajv": "8.20.0",
+    "ajv-formats": "3.0.1",
+    "commander": "15.0.0",
+    "yaml": "2.9.0"
+  },
+  "devDependencies": {
+    "@types/node": "26.4.0",
+    "tsx": "4.23.13",
+    "typescript": "7.0.2",
+    "vitest": "4.1.11"
+  },
+  "pi": {
+    "skills": [
+      "./skills"
+    ]
+  }
+}

package/packs/evidence-backed-market-research.yaml

@@ -0,0 +1,14 @@
+schemaVersion: 1
+id: evidence-backed-market-research
+version: 0.1.0
+name:
+  en: Evidence-backed market research
+  zh-CN: 有证据的市场研究
+description:
+  en: Find authoritative sources before producing a decision-relevant market analysis.
+  zh-CN: 先查找权威来源，再产出支持决策的市场分析。
+atoms: [atom-find-research-sources, atom-analyze-market]
+optionalMcps: []
+ordering:
+  - { before: atom-find-research-sources, after: atom-analyze-market }
+acceptanceTests: [market-research-smoke]

package/packs/localized-research-brief.yaml

@@ -0,0 +1,14 @@
+schemaVersion: 1
+id: localized-research-brief
+version: 0.1.0
+name:
+  en: Localized research brief
+  zh-CN: 本地化研究简报
+description:
+  en: Gather traceable sources and deliver the resulting source brief in another language without losing provenance.
+  zh-CN: 收集可追溯来源，并在不丢失出处的前提下用另一种语言交付来源简报。
+atoms: [atom-find-research-sources, atom-translate-document]
+optionalMcps: []
+ordering:
+  - { before: atom-find-research-sources, after: atom-translate-document }
+acceptanceTests: [localized-research-brief-smoke]

package/packs/safe-skill-evolution.yaml

@@ -0,0 +1,20 @@
+schemaVersion: 1
+id: safe-skill-evolution
+version: 0.2.0
+name:
+  en: Safe Skill evolution
+  zh-CN: 安全 Skill 进化
+description:
+  en: Classify a capability, then govern any proposed lifecycle change through evaluation and rollback gates.
+  zh-CN: 先分类能力，再通过评测与回滚门禁治理生命周期变更。
+atoms: [atom-classify-capability]
+specialSkills: [meta-skill-governor]
+optionalMcps: []
+ordering:
+  - { before: atom-classify-capability, after: meta-skill-governor }
+acceptanceTests: [safe-skill-evolution-protected]
+runtimeState:
+  stateSchema: runtime/schemas/safe-skill-evolution.state.schema.json
+  initialState: runtime/initial/safe-skill-evolution.state.json
+  patchSemantics: json-merge-patch
+  historyPolicy: external-audit-log

package/packs/software-delivery.yaml

@@ -0,0 +1,15 @@
+schemaVersion: 1
+id: software-delivery
+version: 0.1.0
+name:
+  en: Evidence-gated software delivery
+  zh-CN: 证据门禁的软件交付
+description:
+  en: Plan, implement, verify, and security-check a bounded software change.
+  zh-CN: 规划、实现、验证并安全检查一个边界明确的软件修改。
+atoms: [atom-plan-code-change, atom-implement-code-change, atom-audit-source-security]
+optionalMcps: []
+ordering:
+  - { before: atom-plan-code-change, after: atom-implement-code-change }
+  - { before: atom-implement-code-change, after: atom-audit-source-security }
+acceptanceTests: [software-delivery-smoke]

package/paper/references.bib

@@ -0,0 +1,177 @@
+@misc{openai2026agentskills,
+  title = {Agent Skills},
+  author = {{OpenAI}},
+  year = {2026},
+  url = {https://learn.chatgpt.com/docs/build-skills}
+}
+
+@misc{li2026skillsbench,
+  title = {SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks},
+  author = {Xiangyi Li and others},
+  year = {2026},
+  eprint = {2602.12670},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2602.12670}
+}
+
+@misc{kang2026skillret,
+  title = {SkillRet: A Large-Scale Benchmark for Skill Retrieval in LLM Agents},
+  author = {Ryangkyung Kang and Hongcheol Cho and Youngeun Kim},
+  year = {2026},
+  eprint = {2605.05726},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2605.05726}
+}
+
+@misc{zheng2026skillrouter,
+  title = {SkillRouter: Skill Routing for LLM Agents at Scale},
+  author = {YanZhao Zheng and others},
+  year = {2026},
+  eprint = {2603.22455},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.LG},
+  url = {https://arxiv.org/abs/2603.22455}
+}
+
+@misc{liu2026skillswild,
+  title = {How Well Do Agentic Skills Work in the Wild: Benchmarking LLM Skill Usage in Realistic Settings},
+  author = {Yujian Liu and Jiabao Ji and Li An and Tommi Jaakkola and Yang Zhang and Shiyu Chang},
+  year = {2026},
+  eprint = {2604.04323},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.CL},
+  url = {https://arxiv.org/abs/2604.04323}
+}
+
+@misc{gao2026skillweaver,
+  title = {Compositional Skill Routing for LLM Agents: Decompose, Retrieve, and Compose},
+  author = {Xueping Gao},
+  year = {2026},
+  eprint = {2606.18051},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.CL},
+  url = {https://arxiv.org/abs/2606.18051}
+}
+
+@misc{zhao2026skillcomposer,
+  title = {Generative Skill Composition for LLM Agents},
+  author = {Xinyu Zhao and others},
+  year = {2026},
+  eprint = {2606.32025},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.CL},
+  url = {https://arxiv.org/abs/2606.32025}
+}
+
+@misc{liang2026skillnet,
+  title = {SkillNet: Create, Evaluate, and Connect AI Skills},
+  author = {Yuan Liang and others},
+  year = {2026},
+  eprint = {2603.04448},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2603.04448}
+}
+
+@misc{yang2026skillopt,
+  title = {SkillOpt: Executive Strategy for Self-Evolving Agent Skills},
+  author = {Yifan Yang and others},
+  year = {2026},
+  eprint = {2605.23904},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2605.23904}
+}
+
+@misc{tang2026wikiskill,
+  title = {WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution},
+  author = {Liyan Tang and Cyrus Rashtchian and Chun-Sung Ferng and Andrew Tomkins and Da-Cheng Juan and Tu Vu},
+  year = {2026},
+  eprint = {2608.27454},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2608.27454}
+}
+
+@misc{badhe2026skillstate,
+  title = {SKILL.state: Scalable Long-Horizon Agent Skills},
+  author = {Sanket Badhe and Priyanka Tiwari and Jonghyun Chung},
+  year = {2026},
+  eprint = {2608.26263},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2608.26263}
+}
+
+@misc{badhe2026skillsecurity,
+  title = {Agent Skill Security: Threat Models, Attacks, Defenses, and Evaluation},
+  author = {Sanket Badhe and Priyanka Tiwari},
+  year = {2026},
+  eprint = {2607.13987},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.CR},
+  url = {https://arxiv.org/abs/2607.13987}
+}
+
+@misc{schmotz2026skillinject,
+  title = {Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks},
+  author = {David Schmotz and Luca Beurer-Kellner and Sahar Abdelnabi and Maksym Andriushchenko},
+  year = {2026},
+  eprint = {2602.20156},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.CR},
+  url = {https://arxiv.org/abs/2602.20156}
+}
+
+@misc{zhang2026coevoskills,
+  title = {CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification},
+  author = {Hanrong Zhang and others},
+  year = {2026},
+  eprint = {2604.01687},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2604.01687}
+}
+
+@misc{ni2026trace2skill,
+  title = {Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills},
+  author = {Jingwei Ni and others},
+  year = {2026},
+  eprint = {2603.25158},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2603.25158}
+}
+
+@misc{yang2026autoskill,
+  title = {AutoSkill: Experience-Driven Lifelong Learning via Skill Self-Evolution},
+  author = {Yutao Yang and others},
+  year = {2026},
+  eprint = {2603.01145},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2603.01145}
+}
+
+@misc{wang2026skillx,
+  title = {SkillX: Automatically Constructing Skill Knowledge Bases for Agents},
+  author = {Chenxi Wang and others},
+  year = {2026},
+  eprint = {2604.04804},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.CL},
+  url = {https://arxiv.org/abs/2604.04804}
+}
+
+@misc{wang2023voyager,
+  title = {Voyager: An Open-Ended Embodied Agent with Large Language Models},
+  author = {Guanzhi Wang and Yuqi Xie and Yunfan Jiang and Ajay Mandlekar and Chaowei Xiao and Yuke Zhu and Linxi Fan and Anima Anandkumar},
+  year = {2023},
+  eprint = {2305.16291},
+  archivePrefix = {arXiv},
+  primaryClass = {cs.AI},
+  url = {https://arxiv.org/abs/2305.16291}
+}
+