Promote recurring episodes from episodes.md into rules.md after user approval.
npx skills add https://github.com/KaimingWan/oh-my-kiro --skill omk-self-reflect
Please help me install this Agent Skill. Skill: omk-self-reflect Source: https://github.com/KaimingWan/oh-my-kiro/tree/main/skills/omk-self-reflect Command: npx skills add https://github.com/KaimingWan/oh-my-kiro --skill omk-self-reflect
Agent self-learning and rule creation for recurring issues, enabling automated consolidation of recurring mistakes into durable, reusable rules and reducing drift.
Promote a recurring episode to a rule after user approval.
High-intent search queries and answers about installing and using this skill.
Automate agent self-learning by capturing recurring mistakes as episodes, distilling them into formal rules, and writing to rules.md. This reduces agent drift by converting repeated corrections into durable, reusable knowledge after user approval.
Promote an episode to a rule by reading episodes.md, distilling the captured insight into a concise rule, and writing it to knowledge/rules.md. The source episodes are then marked as promoted to prevent duplicate processing.
Episode-based learning is capturing difficult corrections and recurring issues into episodes.md. These episodes are then evaluated, distilled into structured knowledge, and promoted to formal rules to govern future agent flows.
Yes, automated rule promotion integrates with safe, guarded automation. It aligns with top-level instruction files and protected flows, ensuring that only approved rules are written to rules.md while protected files remain untouched.
Convert complex insights into agent rules when recurring patterns or explicit user prompts indicate repeated mistakes. Capturing these difficult corrections as episodes prevents future drift and ensures consistent agent behavior across workflows.
Yes, the three-step agent flow requires user approval before writing to rules.md. The system reads episodes.md, distills a rule, and only commits the knowledge to rules.md after explicit user confirmation, marking the source episode as promoted.
KaimingWanCapture user principles as deduplicated rules in knowledge/rules.md.
0xb11aAudit agent instruction files against source-of-truth code with a five-pass procedure.
Kling0012Create and train AI learning plugins with AgentDB reinforcement algorithms.
kimasplundRank vector-store memories by learned q-values to improve retrieval relevance.
