AgentSkillsCN

go-on

从上次中断的地方继续推进实施流程。每次执行一个原子级别的工作任务,随后干净利落地退出。适用于在功能开发中逐步积累进展。

SKILL.md
--- frontmatter
name: go-on
description:
  Continue the implementation workflow from wherever it was left off. Executes
  ONE atomic unit of work, then exits cleanly. Use to make incremental progress
  on a feature.

Workflow Continuation

Load AGENTS.md to understand the full workflow and current project conventions.

Procedure

  1. Assess current state from external sources:

    • Check for spec (issue tracker, conversation, or docs)
    • Check for implementation plan and task breakdown
    • Check git for branch, commits, uncommitted changes
    • Check filesystem for implementation progress
  2. Identify next action based on state:

    StateAction
    No spec existsWrite spec, update tracking, EXIT
    No plan/tasksWrite plan, present to user, wait for approval, EXIT
    Plan approved but tasks not trackedCreate task entries, EXIT
    Incomplete task existsImplement ONE task, run /quality, commit, mark done, EXIT
    Task in progress, quality failingFix errors until /quality passes, then commit, EXIT
    Code ready but not committedRun /quality, commit, mark task done, EXIT
    All tasks done, no PRCreate PR, EXIT
    PR existsReport completion, EXIT
  3. Execute that ONE action — do NOT ask for permission, just do it

  4. Update tracking (issue tracker, git, task list)

  5. Exit with clear status:

    text
    ## /go-on Status
    
    Previous state: <what was found>
    Action taken: <what was done>
    New state: <what changed>
    Next step: <what should happen next>
    

Key Principles

  • Stateless: Determines state fresh each time from files/git/tracking
  • One atomic unit: Does ONE thing per invocation
  • Idempotent: Safe if interrupted; next invocation re-assesses

Exit Conditions

  • Success: Completed one unit of work
  • Blocked: Cannot proceed (waiting for approval, dependency)
  • Complete: Feature done, PR created
  • Error: Something went wrong, needs human intervention