一键导入
he-implement
Executes active ExecPlans using milestone-driven progress updates, parallel subagents, and evidence-backed verification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Executes active ExecPlans using milestone-driven progress updates, parallel subagents, and evidence-backed verification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Bootstraps a repository for the harness-engineered workflow by creating AGENTS.md, docs/specs, docs/spikes, docs/plans, docs/generated, and baseline tracking files aligned to the PLANS.md contract.
Opens/updates GitHub PRs, checks CI, responds to feedback, and merges after GO using gh CLI with explicit consent gates.
Captures paired browser bug evidence videos using agent-browser (`failure` before fix and `resolution` after fix), stores raw captures in `tmp/artifacts`, and supports minimal manual promotion for durable review evidence. Use for browser/UI bug fixes and release evidence.
Produces a PLANS.md-compliant executable plan from a spec. Use after intake.
Converts fuzzy requests into a concrete initiative spec in docs/specs using a single slug and measurable success criteria. Use at the start of non-trivial work.
Converts post-release outcomes into durable docs and mechanical guardrails, clears generated scratchpads, and archives the initiative plan for reuse.
| name | he-implement |
| description | Executes active ExecPlans using milestone-driven progress updates, parallel subagents, and evidence-backed verification. |
| argument-hint | [slug or docs/plans/active/<slug>-plan.md] |
Execute a PLANS-compliant active plan and keep the plan artifact current.
he-plan when the initiative is ready for executionhe-review or he-verify-release for fixesProgress items and keep living sections current.called_from matches this skill (bash scripts/runbooks/select-runbooks.sh --skill he-implement), but never waive/override anything codified here.git status --short --branch, current branch, worktree detection).bun create next-app, cargo init, framework boilerplate). If so, work directly on main — scaffolding is the project baseline, not a feature. Skip to Phase 1.he-worktree to choose and create the workspace.worktree or branch), branch name, and workspace directory in Decision Log or Revision Notes.docs/plans/active/<slug>-plan.md: Purpose / Big Picture, Context and Orientation, Milestones, Plan of Work, Concrete Steps, and Validation and Acceptance.docs/generated/ when stale:
docs/generated/README.md first (if present) for project-specific expectations.last_updated timestamp when the project uses that convention.Revision Notes.bash scripts/runbooks/select-runbooks.sh --skill he-implement and read any returned runbooks. Apply their additions throughout — they must not waive or override gates codified here.Progress items (P1, P2, ...).Progress items.Revision Notes.Progress items are complete or explicitly deferred.Use subagents aggressively for independent work while keeping integration and plan updates in the main thread.
Subagent return contract — each subagent returns:
Progress item IDsunit or e2e; no mock-only verification)priorityIntegration rules:
Progress items done.he-plan for clarification.After each batch, update docs/plans/active/<slug>-plan.md:
Progress items.Surprises & Discoveries with evidence.Decision Log when approach/scope changes.Outcomes & Retrospective with milestone outcomes/gaps.Artifacts and Notes.Revision Notes with what changed in the plan and why.git add path/to/file ...git add . or git add -A on main/default branch.<type>(<slug>): <milestone summary>For browser UI verification, prefer agent-browser flows and store durable evidence in Artifacts and Notes.
docs/plans/active/<slug>-plan.md with completed progress, evidence, and living sections current.Progress items are completed or explicitly deferredProgress, Surprises & Discoveries, Decision Log, Outcomes & Retrospective, Revision Notes)he-plan for clarification rather than guessing.Surprises & Discoveries, investigate, and escalate if unresolvable.Revision Notes and proceed with available context.| Anti-Pattern | Better Approach |
|---|---|
| Batching all validation at the end | Evidence as you go — verify continuously |
| Using mocks instead of real tests | Unit or e2e only; mocks are a high finding |
| Working on default branch without consent | Workspace isolation gate first |
| Implementing without reading the plan | Plan-driven execution; read before writing |
| Skipping plan updates after each batch | Keep living sections current after every milestone |
Always use interactive question tool at transitions (AskUserQuestion in Claude Code, request_user_input in Codex Plan mode, or equivalent). Offer:
he-review (recommended)he-implementIf running autonomously or no interactive tool is available, continue with he-review and log an Autonomous transition note in Decision Log or Revision Notes.