一键导入
skill-curator
Review skill usage, failures, and repeated procedures, then propose minimal skill changes through the skills provider.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Review skill usage, failures, and repeated procedures, then propose minimal skill changes through the skills provider.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | skill-curator |
| description | Review skill usage, failures, and repeated procedures, then propose minimal skill changes through the skills provider. |
| version | 1.0.0 |
| metadata | {"sloppy":{"tags":["runtime","skills","curation"],"category":"runtime"}} |
Use this skill when repeated session failures, duplicated procedures, stale skills, low-value skill usage, or newly learned workflows suggest that Sloppy's procedural memory should change.
/skills index first. Use view_count, last_viewed_at, tags, category, scope, and supporting_files to identify unused, duplicated, stale, or frequently used skills.skill_view(name) and load supporting files only when the index shows they matter.skill_manage for the smallest change. Prefer session scope for experiments; use workspace/global scope only when the evidence justifies durable procedural memory and the user approves the write.activateSkillVersion topology state instead of making the skill ambient prompt text.view_count means useless when the skill is new or specialized.Confirm the proposed or applied skill change appears under /proposals or /skills, that persistent changes requested approval, and that the updated skill has a clear trigger, procedure, pitfalls, and verification section.
Drive Sloppy persistent session goals through extension-backed metadata instead of hardcoded planning policy.
Use GPT-5.5 and Claude Opus 4.7 ACP sub-agents in an iterative planner/auditor/implementer loop for development tasks.
Coordinate Sloppy child agents through the delegation provider without polling, including parent-side parallel work, result retrieval, and child-session cleanup.
Diagnose failed meta-runtime route dispatch and propose minimal route or endpoint repairs.
Design Sloppy meta-runtime topology changes through ordinary SLOP proposals instead of hardcoded repair policy.
Evaluate meta-runtime topology experiments using explicit evidence and promotion criteria.