continuous-learning

To save debugging insights as reusable skills, you extract non-obvious solutions found during investigation and store them as structured documents. This prevents knowledge loss by preserving hard-won patterns for future agent recall and cross-session learning.

Continuous learning in autonomous agent debugging is the process of extracting reusable patterns from trial-and-error discoveries. It captures non-obvious solutions during debugging sessions or retrospectives to prevent the loss of hard-won knowledge across sessions.

You extract reusable patterns from debugging discoveries by instructing the system to transform the insight into a skill. It writes a pending SKILL.md file, tags it with the extracting agent, and logs trajectory data for auditing and retrieval.

Yes, skill extraction works with end-of-sprint retrospectives. It activates when insights emerge from investigation, experimentation, or trial-and-error, capturing these discoveries as reusable skills to inform future work and improve agent recall.

Extracted skills are stored as pending SKILL.md files in the grimoires/loa/skills-pending directory. The system integrates with trajectory logging and NOTES.md cross-references to avoid duplicates and ensure proper retrieval.