Description
EN. ContextSeek promises automatic progression raw → extracted → knowledge → skill (src/contextseek/domain/stages.py,
src/contextseek/plugs/skills/, domain/skill_executor.py). Build the
distillation pipeline that mines clusters of execution traces / repeated
patterns and distills them into a reusable, parameterized skill item — name,
preconditions, steps, and an executable form the skill_executor can run.
Integrate with the DreamEngine's idle-time consolidation.
Tasks: cluster related traces; LLM-distill a candidate skill with a strict
output schema; validate it (dry-run via skill_executor); store as a skill
stage item with provenance back to the source traces; add an eval that measures
distillation quality; document the lifecycle.
Acceptance: repeated trace patterns yield a validated, executable skill item
linked to its evidence; the executor can invoke it; quality eval is reported.
ZH. ContextSeek 承诺自动晋升 raw → extracted → knowledge → skill
（src/contextseek/domain/stages.py、src/contextseek/plugs/skills/、
domain/skill_executor.py）。请构建蒸馏流水线：从执行 trace 的聚类/重复模式中挖掘，
蒸馏成可复用、可参数化的 skill 条目——名称、前置条件、步骤，以及 skill_executor
可执行的形式。与 DreamEngine 的闲时巩固集成。
任务： 对相关 trace 聚类；用严格输出 schema 让 LLM 蒸馏候选 skill；校验它（经
skill_executor dry-run）；存为 skill 阶段条目并保留回溯到源 trace 的 provenance；
加一个衡量蒸馏质量的 eval；文档化生命周期。
验收： 重复 trace 模式能产出经校验、可执行、链回证据的 skill 条目；executor 能调
用它；给出质量 eval 报告。
Description
EN. ContextSeek promises automatic progression
raw → extracted → knowledge → skill(src/contextseek/domain/stages.py,src/contextseek/plugs/skills/,domain/skill_executor.py). Build thedistillation pipeline that mines clusters of execution traces / repeated
patterns and distills them into a reusable, parameterized skill item — name,
preconditions, steps, and an executable form the
skill_executorcan run.Integrate with the DreamEngine's idle-time consolidation.
Tasks: cluster related traces; LLM-distill a candidate skill with a strict
output schema; validate it (dry-run via
skill_executor); store as askillstage item with provenance back to the source traces; add an eval that measures
distillation quality; document the lifecycle.
Acceptance: repeated trace patterns yield a validated, executable skill item
linked to its evidence; the executor can invoke it; quality eval is reported.
ZH. ContextSeek 承诺自动晋升
raw → extracted → knowledge → skill（
src/contextseek/domain/stages.py、src/contextseek/plugs/skills/、domain/skill_executor.py）。请构建蒸馏流水线：从执行 trace 的聚类/重复模式中挖掘，蒸馏成可复用、可参数化的 skill 条目——名称、前置条件、步骤，以及
skill_executor可执行的形式。与 DreamEngine 的闲时巩固集成。
任务： 对相关 trace 聚类；用严格输出 schema 让 LLM 蒸馏候选 skill；校验它（经
skill_executordry-run）；存为skill阶段条目并保留回溯到源 trace 的 provenance；加一个衡量蒸馏质量的 eval；文档化生命周期。
验收： 重复 trace 模式能产出经校验、可执行、链回证据的 skill 条目；executor 能调
用它；给出质量 eval 报告。