AgentSkillsCN

change-risk-rollback

每当有变更拟用于发布或部署时，运用此技能评估运营风险，识别潜在故障模式，并确保回滚路径清晰且切实可行。

SKILL.md
--- frontmatter
name: change-risk-rollback
description: >
  Use this skill whenever a change is intended for release or deployment to
  assess operational risk, identify failure modes, and ensure rollback paths are
  explicit and realistic.

Change Risk & Rollback Assessment

Intent

Assess the operational risk of changes and ensure rollback options are explicit, tested where possible, and aligned to traceability practices.

When to Use

  • •Before releasing or deploying significant changes
  • •When touching critical paths, data handling, auth, or infrastructure
  • •When introducing migrations or behavioural changes with user impact

Precondition Failure Signal

  • •No clear failure modes or rollback strategy exists
  • •Rollback depends on guesswork or mutable deployments
  • •Risk is described qualitatively without concrete triggers/mitigations
  • •Deployment plan lacks version/tag traceability

Postcondition Success Signal

  • •Key risks and failure modes are identified and bounded
  • •Rollback options are explicit and traceable to versions/tags
  • •Mitigations are aligned to quality gates and environment traceability
  • •Decision to proceed is reviewable and justified

Process

  1. •Source Review: Inspect the proposed changes, deployment environment, and current telemetry/monitoring coverage.
  2. •Implementation: Identify potential failure modes and define the corresponding rollback or remediation steps.
  3. •Verification: Conduct a walkthrough or simulation of the rollback plan to ensure it is executable and effective.
  4. •Documentation: Record the risk assessment and rollback plan in the change request or as an ADR for high-risk architectural changes.
  5. •Review: Release Manager/SRE and Tech Lead must validate the feasibility and completeness of the rollback strategy.

Example Test / Validation

  • •For a change, identify at least one failure mode and define a rollback path
  • •Verify rollback candidate versions/tags can be identified from traceability records
  • •Confirm mitigations are in place for highest-likelihood/highest-impact failures

Common Red Flags / Guardrail Violations

  • •“We can just hotfix it”
  • •Relying on manual steps without documentation
  • •Deploying without tags, making rollback ambiguous
  • •Treating rollback as “restore from backup” without feasibility validation

Recommended Review Personas

  • •Tech Lead – validates risk framing and scope discipline
  • •Platform/DevOps Engineer – validates deployment/rollback feasibility
  • •Domain Expert – validates business/functional failure impacts
  • •Security Reviewer – validates integrity risks where applicable

Skill Priority

P0 – Safety & Integrity

Conflict Resolution Rules

  • •Overrides delivery pressure when rollback/risk is unclear
  • •Requires traceability (versions/tags); no rollback plan without provenance
  • •If risk cannot be bounded, escalate and defer release

Conceptual Dependencies

  • •incremental-change-impact
  • •release-tagging-plan
  • •environment-traceability

Classification

Core
Governance

Notes

Rollback is not a plan if it cannot be executed from recorded state.