D7: Pattern Recognition (10 points)

Purpose: Ensure the skill activates when needed via description keywords and trigger conditions.

Scoring:

Requirements

• Description must include domain keywords
• Trigger scenarios in the description or a “When to Apply” section
• Example: “Use when writing BDD tests, feature files, Gherkin scenarios…”

The best description = exhaustive trigger list + concrete examples.

Discriminativeness (diagnostic signal)

A high-quality description reduces false positives by anchoring the skill to specific contexts:

1. Negative anchor — explicitly states when NOT to activate
   (e.g., Does not apply, SKIP when, Not for, Exclude, DO NOT trigger, not intended for)

2. Workflow anchor — trigger tied to a concrete artifact or action
   (e.g., references file, PR, commit, test, config, pipeline, migration)

This is a diagnostic signal only — it does not affect the numeric score in the current iteration.

Academic References

@article{zhang2025agentrouter,
  title         = {AgentRouter: A Knowledge-Graph-Guided LLM Router for Collaborative Multi-Agent Question Answering},
  author        = {Zhang and others},
  year          = {2025},
  journal       = {arXiv preprint arXiv:2510.05445},
  eprint        = {2510.05445},
  archivePrefix = {arXiv},
  url           = {<https://arxiv.org/abs/2510.05445}>
}

@article{wang2026aco,
  title         = {Efficient and Interpretable Multi-Agent LLM Routing via Ant Colony Optimization},
  author        = {Wang and others},
  year          = {2026},
  journal       = {arXiv preprint arXiv:2603.12933},
  eprint        = {2603.12933},
  archivePrefix = {arXiv},
  url           = {<https://arxiv.org/abs/2603.12933}>
}

@article{yehudai2025survey,
  title         = {Survey on Evaluation of LLM-Based Agents},
  author        = {A. Yehudai and L. Eden and A. Li and G. Uziel and Y. Zhao and R. Bar-Haim and A. Cohan and M. Shmueli-Scheuer},
  year          = {2025},
  journal       = {arXiv preprint arXiv:2503.16416},
  eprint        = {2503.16416},
  archivePrefix = {arXiv},
  url           = {<https://arxiv.org/abs/2503.16416>}
}

@inproceedings{chen2024agentpoison,
  title         = {AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases},
  author        = {Z. Chen and Z. Xiang and C. Xiao and D. Song and B. Li},
  year          = {2024},
  booktitle     = {Advances in Neural Information Processing Systems (NeurIPS 2024)},
  url           = {<https://proceedings.neurips.cc/paper_files/paper/2024/hash/eb113910e9c3f6242541c1652e30dfd6-Abstract-Conference.html>}
}