🛰️ Daily AI Frontier
‹ back to 2026-08-13

Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction

Research LLM Agents

Ranking

Overall 85
Content 95
Popularity 63

Observed public metrics from 1 member.

Merged summary

TL;DR - DARC is a diagnosis-guided recovery framework that identifies task-specific agent failure modes before deployment, then selectively applies compatible correction strategies. It improves task performance while reducing environment steps or retrieval costs compared with base agents and broad recovery playbooks.

  • Profiles development-set failures to determine which recovery interventions are admissible.
  • Prunes mismatched strategies and freezes a verifier-selected success-cost policy for test-time use.
  • Produces task-specific recovery mechanisms for action validity, procedural failures, and strict formatting.
  • Demonstrated improvements across ALFWorld, AppWorld, and XBRL Finance.

Sources (1)

Diagnosis Before Recovery: Turning Agent Failures into Selective Self-Correction

arXiv cs.CL Pan Wang, Yihao Hu, Hang Wang, Zirui Lv, Xin Zhang, Jianshe Li, Jiang-Ming Yang, Wei Wu, Yongqi Tong 2026-08-12 arXiv:2608.11772
Public signals Hugging Face upvotes 2
Providers: Hugging Face · Upvotes 2 OpenAlex · N/A Publisher · N/A Semantic Scholar · N/A X · N/A Fetched 2026-09-12 14:27:37.912749 UTC

TL;DR - DARC is a diagnosis-guided recovery framework that identifies task-specific agent failure modes before deployment, then selectively applies compatible correction strategies. It improves task performance while reducing environment steps or retrieval costs compared with base agents and broad recovery playbooks.

  • Profiles development-set failures to determine which recovery interventions are admissible.
  • Prunes mismatched strategies and freezes a verifier-selected success-cost policy for test-time use.
  • Produces task-specific recovery mechanisms for action validity, procedural failures, and strict formatting.
  • Demonstrated improvements across ALFWorld, AppWorld, and XBRL Finance.
item →