From builders-dont-cry
Use when the user explicitly invokes /autoloop to automate an already-approved plan through /execute, /verify, and /ship with bounded retries and handoff on repeated failure.
How this skill is triggered — by the user, by Claude, or both
Slash command
/builders-dont-cry:autoloopThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
If builder says `别走流程`, `直接做`, `简单点`, or `不走地图`, do not run `/autoloop`. Acknowledge once and do the smallest direct action.
If builder says 别走流程, 直接做, 简单点, or 不走地图, do not run /autoloop. Acknowledge once and do the smallest direct action.
Exception: the escape hatch does not bypass safety gates for destructive changes, secret access, public upload, deploy, database migration, force push, or irreversible writes.
/autoloop is an explicit automation wrapper for mature, already-approved plans.
It may run:
/execute -> /verify -> /ship
It must not create a plan, expand scope, choose new dependencies, perform destructive actions, push, open a PR, or publish unless the active plan and /ship both say that action is in scope and builder explicitly approves it.
Before doing anything else:
bash ${BDC_HOME:-$HOME/.bdc}/scripts/active-plan-resolver.sh --single
If any precondition fails, stop and report the exact missing condition. Do not downgrade into ad-hoc /execute.
/execute on the next unchecked step./execute passes, run /verify on the finished work./verify returns APPROVE, continue until the next boundary or until no unchecked steps remain./verify returns REQUEST_CHANGES, treat it as failure 1 and run one correction loop through /execute./handoff./ship for local-first go/no-go. Do not push, publish, or open a PR unless /ship asks for it and builder confirms.| Event | Count |
|---|---|
/execute verify failure | 1 |
/verify REQUEST_CHANGES | 1 |
/ship NO-GO with actionable fix | 1 |
| Same stage fails again after one correction loop | stop |
Two accumulated failures in one /autoloop invocation require:
/handoff or write the equivalent handoff if the handoff skill is unavailable/autoloop. Because: automation must not hide irreversible decisions./autoloop is a local execution wrapper, not a release authority./execute scope checks define the atomic boundary./autoloop starts after planning is frozen.Keep a compact running ledger:
Autoloop:
- Plan: <path>
- Current stage: execute | verify | ship | stopped
- Passed: <count>
- Failures: <count>/2
- Last evidence: <command or verdict>
- Next action: <continue | stop | handoff | ship>
active-plan-resolver.sh --single./execute produced fresh verify evidence before /verify./verify returned APPROVE before /ship./ship produced GO/NO-GO and rollback/restore notes before any release action./autoloop starts from a fuzzy idea or unreviewed plan.REQUEST_CHANGES as a warning instead of a failure.| Rationalization | Response |
|---|---|
| "The plan is probably approved because it exists." | Do not assume. Incident: standards/debugging.md 2026-04-04 records written rules being ignored until hard gates enforced them; approval must be explicit. |
| "One more retry might fix it." | Stop after the budget. Incident: large health baseline runs require durable handoff state instead of unbounded continuation. |
${BDC_HOME:-$HOME/.bdc}/standards/roadmap.md${BDC_HOME:-$HOME/.bdc}/skills/execute/SKILL.md${BDC_HOME:-$HOME/.bdc}/skills/verify/SKILL.md${BDC_HOME:-$HOME/.bdc}/skills/ship/SKILL.md${BDC_HOME:-$HOME/.bdc}/skills/handoff/SKILL.mdCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub 0xmariowu/builders-dont-cry