{{ skill.trigger }}
{{ skill.body }}When this skill matches a task strongly, the macro below is executed deterministically — zero LLM tokens, no model latency between steps. On any error the agent falls back to the regular ReAct loop.
Confidence: {{ "%.2f"|format(macro_conf) }} · Steps: {{ macro_steps|length }} · Distilled from: {{ macro_episodes|length }} successful episodes
{{ st.tool }}
{{ st.args_json }}
Generated as an alternative to a failing skill: {{ skill.parent_skills[0][:8] }}. This skill competes with the parent for retrieval; if it outperforms, the parent is eventually retired.
This skill's fitness is in the uncertain zone (~{{ "%.2f"|format(skill.fitness_mean) }}). The dreamer wrote concrete tasks the agent could try — running them feeds real evidence into the Bayesian fitness so the skill is either promoted or retired.