AgentSkillsCN

system objective state

当您需要跨会话管理任务状态、在压缩后继续处理,或针对超出上下文限制的多步骤任务时,可使用此工具。

SKILL.md
--- frontmatter
name: system objective state
description: Use when you need to manage task state across sessions, after compaction, or for multi-step tasks that exceed context limits.

Purpose

Persist and recover task state using the Objective Graph to survive context loss and ensure continuity across sessions.

Scope

Covers the Objective Graph schema, state management lifecycle, and operational findings promotion. Does not cover governance rules (see system-governance) or context mechanics (see system-context-mechanics).

Minimal Path

  1. At session start, check if docs/work/objective-graph.yaml exists and load active objectives.
  2. For complex tasks, create an objective with id, goal, status, and next_actions.
  3. Update the objective as work progresses, recording blockers and findings.
  4. Before closing an objective, promote any operational_findings to canonical docs.
  5. Mark objective done only when all acceptance criteria are met.

Validation

  • Objective Graph parses as valid YAML matching the schema.
  • Active objectives have clear next_actions defined.
  • Closed objectives have empty or promoted operational_findings.

Failure Modes

  • Starting work without checking existing objective state (context loss).
  • Leaving objectives open with stale next_actions.
  • Closing objectives without promoting findings (knowledge loss).

References

  • docs/system/model/objective-graph.md: Schema and semantics
  • docs/system/procedure/maintaining-objective-graph.md: Lifecycle procedure
  • docs/system/procedure/promoting-operational-findings.md: Findings promotion
  • docs/system/procedure/objective-graph-realignment.md: Recovery from drift