engineering-loop
Implement a repository code or configuration change through reproduction, focused checks, review, and evidence. Use for end-to-end features and fixes; not for explanation-only, review-only, research, content publishing, or production operations.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- engineering-loop
- description
- Implement a repository code or configuration change through reproduction, focused checks, review, and evidence. Use for end-to-end features and fixes; not for explanation-only, review-only, research, content publishing, or production operations.
02 资源文件 6
- 参考文档agents/openai.yaml230 B
- 参考文档references/evidence-example.md2.0 KB
- 参考文档references/hard-debugging.md1.8 KB
- 参考文档references/independent-review.md3.8 KB
- 参考文档references/loop-contract.md4.6 KB
- 参考文档references/loop-policy.md2.8 KB
03 SKILL.md 正文
Engineering Loop
Deliver a verified change with minimal supervision and proportionate checking. Start with one agent. Delegate only when authorized and independent work or review coverage justifies the overhead. Follow applicable repository and team requirements; this skill adds no ticketing or mandatory-agent process.
Establish the baseline
- Read applicable
AGENTS.mdfiles and repository documentation. - Record the requested behavior, constraints, and observable done conditions.
- Inspect the branch and working tree; preserve unrelated user-owned changes.
- Before changing shared behavior, trace affected callers, consumers, and tests using targeted search or an available code graph. Expand along relevant paths.
- Identify repository-native validation and run the smallest safe baseline that separates pre-existing failures from task regressions.
Resolve routine choices from evidence. Ask only for missing authority or a material decision: ambiguous acceptance, incompatible public behavior, or destructive/data-changing side effects not already authorized. Continue independent authorized work while awaiting an answer.
Load only the needed reference:
- Long, cross-stack, or high-risk work: loop-contract.md.
- Repeated iterations or explicit budgets: loop-policy.md.
- Intermittent, performance, or unresolved defects: hard-debugging.md.
- Requested or risk-justified separate review: independent-review.md.
Run the loop
- Reproduce the defect, or capture the current behavior for a feature.
- Choose the smallest coherent change and the evidence that will prove it.
- Add a failing regression test first when practical.
- Implement one bounded change and run the narrowest relevant check.
- If an unplanned path is necessary, record why and update the scope and checks.
- Classify failures as product, test, environment, assumption, or external blocker. Do not change product code to mask environment failures.
- Run affected consumer checks and repository-required broader checks.
- Review the complete diff first for task-contract compliance, then for code quality, regressions, security, and maintainability.
- Fix consequential findings and rerun checks affected by those fixes.
Once acceptance evidence and required checks pass, broaden or repeat checks only for new changes, failures, or unresolved concerns. Add tests when they prove behavior or a meaningful invariant, not merely mirror a low-impact edit.
If no test harness exists, use a small repeatable script or documented manual check with inputs, expected behavior, and observed results. Record missing automated coverage; a build alone does not prove runtime behavior. Do not add a large testing framework solely to satisfy the loop.
Keep a compact ledger of facts, changed files, commands, outcomes, and the next decision instead of full logs.
Stop conditions
- After two identical failures, stop blind retries and reclassify the cause. Resume only with changed relevant state or an evidence-producing probe.
- If discovery or review repeats without new evidence, checkpoint the diff, unresolved issue, and next useful probe. Do not restart the same cycle.
- Stop and report a blocker when progress requires missing authority, secrets, unavailable infrastructure, an unauthorized destructive action, or an unresolved material product decision.
- When an explicit loop budget is exhausted, stop with the latest judge evidence instead of silently expanding the budget. Without an explicit budget, continue only while a bounded next action can produce new evidence; productive iterations have no arbitrary fixed count.
- Never make a failing check pass by weakening assertions, deleting coverage, hiding errors, or silently changing acceptance criteria.
- Do not query production, deploy, migrate, merge, or publish unless the user explicitly authorizes that action.
Finish with evidence
Report changed behavior and files, regression or acceptance proof, exact check commands and outcomes, review disposition, and remaining limitations. Include the final reviewed revision when independent review was used. Never label blocked validation or unresolved consequential findings as complete. See evidence-example.md for a filled-in handoff.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.