菜单
Maintain the live agent context surface: bootstrap hygiene, memory routing, conversation continuity, and context-health monitoring.
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。
jarvOS personal AI operating system — project management, ontology context, governance, and behavioral rules for your assistant.
Validate, audit, and maintain scheduled agent automation without unsafe direct state edits.
Wire new rules, policies, and behavioral constraints into the correct governance file with an explicit enforcement assessment.
Plan-first workflow for non-trivial work: define the goal, create or reuse a tracker issue, package context, execute on an issue-named branch when code changes, and verify completion with evidence.
基于 SOC 职业分类
| name | context-management |
| description | Maintain the live agent context surface: bootstrap hygiene, memory routing, conversation continuity, and context-health monitoring. |
Use this skill when editing always-loaded files, deciding where durable knowledge belongs, diagnosing context bloat, or preserving continuity across compaction.
The workflow is complete only when:
| Lane | Use when | Output |
|---|---|---|
| Bootstrap hygiene | Root instruction files are too large, stale, or mixed-purpose | Smaller root files with pointers to deeper docs/skills |
| Memory routing | A fact, preference, lesson, decision, or project state needs a home | Correct durable memory/document location |
| Conversation continuity | Compaction, transcript recovery, or session handoff is the problem | Recovery summary and source-of-truth refresh |
| Monitoring/maintenance | Watchdogs, context budgets, or trend reports show drift | Diagnosis, targeted cleanup, and verification |
Keep root instructions short. They should contain durable rules, routing pointers, and triggers. Move long procedures to skills, references, or docs.
Before deleting content, diagnose why it is no longer needed. Watchdog output is evidence, not an automatic pruning command.
After editing:
