Paper Citation Record · LEDGER

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents

As of 20 August 2026, this Paper Citation Record lists 43 of 43 outbound references and 0 inbound Pith citation observations for arXiv:2608.12851.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2608.12851 v1

Coverage vector

measured 43 of 43 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-15T22:04:21.557975Z

measured 43 of 43 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-20T06:33:59.587034+00:00

measured 0 of 0 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: cited_works

Reference resolution

43 of 43 outbound references displayed

verified exact0
verified fuzzy3
unresolved37
parse uncertain2
malformed identifier0
metadata mismatch1

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 29f65e75-4bb1-4a2a-a451-b13130a6f68e · outbound

This paper cites arXiv preprint arXiv:2508.19005 , year=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents arXiv preprint arXiv:2508.19005 , year=

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.378849Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ec77adfb-9972-4ccb-a22a-78dad98c22c1 · outbound

This paper cites AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents

Reference 2

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.383625Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 935a8678-c07a-4d8b-b237-262c2f006817 · outbound

This paper cites Agent Hacks Agent: Autoresearch for Production-Agent Red-Teaming.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Agent Hacks Agent: Autoresearch for Production-Agent Red-Teaming

Reference 3

Resolution: metadata mismatch; local_arxiv, observed 2026-08-15T22:04:22.185660Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-20T06:33:59.587034+00:00.

Observation 76cbcabd-9f83-48db-a73f-a21e62dc0256 · outbound

This paper cites SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.392228Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 876f65f3-da93-4952-8737-0afc422e1285 · outbound

This paper cites MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.396643Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d220a01c-7455-41da-9c82-75d80a92db46 · outbound

This paper cites SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.400883Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation df8e359c-7003-4c40-98d7-6c2998bf653e · outbound

This paper cites SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation on Real-World Tasks.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation on Real-World Tasks

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.405688Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bdb23f5b-a4be-4ccd-b10b-eea57cebe778 · outbound

This paper cites EvoSkill: Automated Skill Discovery for Multi-Agent Systems.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents EvoSkill: Automated Skill Discovery for Multi-Agent Systems

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.410437Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fcdfd702-6ebe-410f-a877-d0541f5c241a · outbound

This paper cites arXiv preprint arXiv:2603.01145 , year=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents arXiv preprint arXiv:2603.01145 , year=

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.415809Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d79c8954-aa84-4978-983c-9487315b1b52 · outbound

This paper cites SkillClaw: Let Skills Evolve Collectively with Agentic Evolver.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillClaw: Let Skills Evolve Collectively with Agentic Evolver

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.419240Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bd43d7c1-218b-43db-8ebe-d3dec77d60ca · outbound

This paper cites SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.422994Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e48a6af3-d109-4207-95c3-e8ac4f2368b3 · outbound

This paper cites SkillOpt: Executive Strategy for Self-Evolving Agent Skills.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillOpt: Executive Strategy for Self-Evolving Agent Skills

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.426994Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 888c1d0d-3d32-4b03-8963-aa768b317ebb · outbound

This paper cites CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.430591Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e04a4d99-168b-44f1-8fa4-cdf234c2eb4e · outbound

This paper cites MUSE-Autoskill: Self-Evolving Agents via Skill Creation, Memory, Management, and Evaluation.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents MUSE-Autoskill: Self-Evolving Agents via Skill Creation, Memory, Management, and Evaluation

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.434517Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 281883ba-7d67-46c1-85c3-a9fce39469a5 · outbound

This paper cites arXiv preprint arXiv:2606.14239 , year=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents arXiv preprint arXiv:2606.14239 , year=

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.438406Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 738c0697-8011-4f21-8458-62b7560dc474 · outbound

This paper cites Anything2Skill: Compiling External Knowledge into Reusable Skills for Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Anything2Skill: Compiling External Knowledge into Reusable Skills for Agents

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.441790Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 14673bdc-fca2-4285-ad00-8f26cb291e58 · outbound

This paper cites Socratic-SWE: Self-Evolving Coding Agents via Trace-Derived Agent Skills.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Socratic-SWE: Self-Evolving Coding Agents via Trace-Derived Agent Skills

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.445587Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c30e20be-b8a3-4229-a617-cbb8cb9f48e4 · outbound

This paper cites CODESKILL: Learning Self-Evolving Skills for Coding Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents CODESKILL: Learning Self-Evolving Skills for Coding Agents

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.449368Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1b5cd27c-d2e2-4cf1-adc6-5bfa34ecbfc3 · outbound

This paper cites SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.453477Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c01c65cf-4d9c-4ebd-b850-5a3b098d4d6d · outbound

This paper cites SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.458110Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 69551db2-91e0-4a22-8bc4-bef855be46b9 · outbound

This paper cites SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale

Reference 21

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.462466Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3d26f46c-cfd0-428b-84bf-360b6c1bba6a · outbound

This paper cites arXiv preprint arXiv:2603.15401 , year=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents arXiv preprint arXiv:2603.15401 , year=

Reference 22

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.466960Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f8598ecd-c8af-4450-9371-1e1ad8ad53f6 · outbound

This paper cites arXiv preprint arXiv:2509.26354 , year=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents arXiv preprint arXiv:2509.26354 , year=

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.470772Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bdff057e-a432-48f4-9f7b-04075dea8db2 · outbound

This paper cites Findings of the Association for Computational Linguistics: ACL 2026 , pages=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Findings of the Association for Computational Linguistics: ACL 2026 , pages=

Reference 24

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.475080Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 497ae427-5fc2-4714-a548-b796b28f2e42 · outbound

This paper cites MemEvoBench: Benchmarking Safety Risks from Memory Misevolution in LLM Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents MemEvoBench: Benchmarking Safety Risks from Memory Misevolution in LLM Agents

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.480294Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 24a07787-8ddd-452f-b491-99eda92aa8ef · outbound

This paper cites TAME: A Trustworthy Test-Time Evolution of Agent Memory with Systematic Benchmarking.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents TAME: A Trustworthy Test-Time Evolution of Agent Memory with Systematic Benchmarking

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.484934Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 63f7c422-8b03-48ac-a043-32ec31a2da56 · outbound

This paper cites Do Self-Evolving Agents Forget? Capability Degradation and Preservation in Lifelong LLM Agent Adaptation.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Do Self-Evolving Agents Forget? Capability Degradation and Preservation in Lifelong LLM Agent Adaptation

Reference 27

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.490117Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 680a5b1c-2f90-46fb-8c6d-713cf7fd96ce · outbound

This paper cites Safety in Self-Evolving LLM Agent Systems: Threats, Amplification, and Case Studies.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Safety in Self-Evolving LLM Agent Systems: Threats, Amplification, and Case Studies

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.495032Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 72b7dbfa-0962-4a7c-b718-16dabb40febf · outbound

This paper cites Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.499359Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4336d6a3-9603-49fe-bbef-e389e1f69abe · outbound

This paper cites Maddison and Tatsunori Hashimoto , booktitle=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Maddison and Tatsunori Hashimoto , booktitle=

Reference 30

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.503867Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f8b5e573-d84a-4b8f-aeba-96ebc95d6ce6 · outbound

This paper cites Advances in Neural Information Processing Systems , volume=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Advances in Neural Information Processing Systems , volume=

Reference 31

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.507941Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 39b576ad-2b05-4541-be3d-5ac1cd238982 · outbound

This paper cites AgentHarm: A Benchmark for Measuring Harmfulness of.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents AgentHarm: A Benchmark for Measuring Harmfulness of

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.512102Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4e989116-3886-4319-8075-c63315b5defd · outbound

This paper cites Agent Security Bench (.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Agent Security Bench (

Reference 33

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T22:04:22.331832Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-20T06:33:59.587034+00:00.

Observation 09492616-ed78-40e5-bd47-5450c1ea9829 · outbound

This paper cites Agent-SafetyBench: Evaluating the Safety of LLM Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Agent-SafetyBench: Evaluating the Safety of LLM Agents

Reference 34

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.522712Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7169a9c4-03c8-4ea2-a9f8-c9de1163e4ad · outbound

This paper cites The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

Reference 35

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T22:04:22.316371Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-20T06:33:59.587034+00:00.

Observation 3bba6d3f-8246-4d6b-b0b8-65a4946b3351 · outbound

This paper cites OpenAgentSafety: A Comprehensive Framework For Evaluating Real-World.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents OpenAgentSafety: A Comprehensive Framework For Evaluating Real-World

Reference 36

Resolution: verified fuzzy; raw_fallback, observed 2026-08-15T22:04:22.301512Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-20T06:33:59.587034+00:00.

Observation a4dcdecd-02a2-4603-b3f4-d029c6ce3893 · outbound

This paper cites DecodingTrust-Agent Platform (DTap): A Controllable and Interactive Red-Teaming Platform for AI Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents DecodingTrust-Agent Platform (DTap): A Controllable and Interactive Red-Teaming Platform for AI Agents

Reference 37

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.536137Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9be6a542-8daa-471e-8f38-2277f3d7ba5f · outbound

This paper cites AgentSecBench: Measuring Prompt Injection, Privacy Leakage, and Tool-Use Integrity in LLM Agents.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents AgentSecBench: Measuring Prompt Injection, Privacy Leakage, and Tool-Use Integrity in LLM Agents

Reference 38

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.540541Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 92c46af7-e3e8-48ae-ae63-90c5d746ab2d · outbound

This paper cites ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection

Reference 39

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.544774Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e5761892-556e-493e-92af-792646a6a01e · outbound

This paper cites NetInjectBench: Benchmarking Indirect Prompt Injection in Tool-Using Large Language Model Agents for Network Operations.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents NetInjectBench: Benchmarking Indirect Prompt Injection in Tool-Using Large Language Model Agents for Network Operations

Reference 40

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.547978Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f54ec0e1-6d25-4c77-8ab9-487abc9fc46f · outbound

This paper cites an unresolved cited work.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Unresolved cited work

Reference 41

Resolution: parse uncertain; no resolver link, observed 2026-08-15T22:04:21.551481Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 15ceeade-e2c6-465f-a136-9717f20bea27 · outbound

This paper cites an unresolved cited work.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Unresolved cited work

Reference 42

Resolution: parse uncertain; no resolver link, observed 2026-08-15T22:04:21.554879Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ab08e74f-48b2-4c88-9c48-3a7617ec1857 · outbound

This paper cites Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries.

Practice Makes Unsafe: Skill Misevolution in Self-Improving LLM Agents Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries

Reference 43

Resolution: unresolved; no resolver link, observed 2026-08-15T22:04:21.557975Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

No inbound Pith citation observations are available.