helpin-ai/agent-runtime/approval-protocol
Claude Code and agent skill by helpin-ai/agent-runtime
- community
Checking that the SKILL.md file is still in the repository.
Inline approval and review-checkpoint contract for planner-style runs.
Install
npx vskill@latest install helpin-ai/agent-runtime/approval-protocolSKILL.md
Approval requests happen inline in the same chat.
- When the PRD is ready for approval, call
publish_prd_draft, then callrequest_approvalwithphase="prd", then stop. - When the task plan is ready for approval, call
publish_task_plan, then callrequest_approvalwithphase="tasks", then stop. - When the task planning document is ready for approval, call
publish_task_plan_doc, then callrequest_approvalwithphase="task_doc", then stop. - An approval request must refer to a preview artifact published in that same turn. When there are multiple previews in the same turn, include
preview_panel_keyon the approval request so it binds to the correct preview. - The human may approve or request changes with a normal chat reply. Do not tell them to use a separate approval state, button, or workflow.
- Treat
request_approvalas the last action in that turn. Do not call more tools after it in the same turn. Do not add "what would you like to do next" or restate approval options after requesting approval. - After explicit PRD approval, continue automatically into task planning in the same run. Do not ask whether you should proceed to tasks unless the human asked to change scope.
- After PRD approval is persisted, your next turn must continue into task planning. Either ask the next blocking questions with
request_user_inputor publish the task plan preview. Do not complete the run immediately after PRD approval.
Required Approval Tool
When you are ready for approval, call request_approval with this shape:
Do not ask for approval in any other format.
Request Changes
When the human requests changes, incorporate that feedback into the current phase output, revise it, republish the full replacement preview for that phase, and emit a fresh approval block again when ready.
For phase="task_doc", this means: revise the planning document markdown, publish the full replacement draft with publish_task_plan_doc, then call request_approval again with phase="task_doc" when the revision is ready.
Do not end the run with a prose-only acknowledgement after change feedback. The next step after change feedback is to revise the active artifact, ask blocking questions with request_user_input, or request another approval request when the revision is ready.
Only treat the phase as approved when the human gives a clear, explicit approval. If the human asks for changes, raises concerns, asks follow-up questions, or gives mixed or ambiguous feedback, treat that as not approved and revise instead of moving forward.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckinginternal/skills/system/approval_protocol/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/helpin-ai/agent-runtime/approval-protocol)