AgentSkillsCN

runner-liveness-guardrails

在升级过程中，保持本地Runner的全天候运行，并确保控制目标的精准性。在更改Runner的启停逻辑、启动器API、Runner凭据签发、端口检测，或多实例运行行为时使用。

SKILL.md
--- frontmatter
name: runner-liveness-guardrails
description: Preserve 24/7 local runner operation and control-target accuracy during upgrades. Use when changing runner start/stop, launcher APIs, runner credential issuance, port detection, or multi-instance behavior.

Runner Liveness Guardrails

Preserve runtime independence

  • •Keep runner alive without active owner browser session.
  • •Keep long-lived runtime auth scoped to runner credential, not web session bearer.

Enforce control-target accuracy

  • •Treat runner as running for a selected agent only if launcher status matches selected agentId.
  • •Before start, preflight selected port and block if occupied by another agent.
  • •Before stop, preflight selected port and block stop for agent mismatch.
  • •Preserve explicit user-selected port; do not force-switch silently.

Preserve observability

  • •Keep per-port log separation and metadata stamping (instanceId/port/pid/agentId).
  • •Keep communication/full-trace logging behavior intact when touching engine boundaries.
  • •Keep runner<->agent communication semantics (action schema + tx_feedback logging) stable; pair with runner-communication-protocol-guardrails when changing protocol fields.

Required scenario checks

  • •Execute 2-port/2-agent matrix (4391/4392, agents A/B).
  • •Verify cross-select start/stop mismatch blocking.
  • •Verify log files remain separated by port and stamped with instance metadata.