agent-tool-call.workflow — Claude Code prompt
agent-tool-call.workflow — a prompt for building a workflow diagram where an AI agent calls tools with human approval gates.
Apache-2.0 ★ 98,599 updated 2026-10-25
.claude/commands/agent-tool-call.workflow.mdWhat agent-tool-call.workflow does
This prompt structures a workflow for tool invocation by an AI agent, with human sign-off at each gate. A user request reaches the chat surface and planner; a router decides whether approval is needed. Approved invocations run against an external provider; denied ones block and can re-request. Tool results and context are logged for inspection.
The prompt targets a typed JSON diagram written with Archify, a skill from the repository. You define block IDs and layout, making the approval decision, denial path, and external call legible. Validation is available locally via Archify CLI; the final candidate goes to benchmark-candidate.json at the repository root.
What it helps with
- Visualize the approval gate for agent tool invocation
- Show denial path and re-request flow
- Record execution results and context for debugging
- Generate a clean JSON diagram for framework integration
Example prompts for the agent
- Build a workflow diagram showing agent tool calls with approval gates and denial handling
- Create a JSON diagram where requests route through a planner, get approved or denied, then execute or retry
- Generate an Archify diagram of tool-call flow with tracing and external provider calls
How to install agent-tool-call.workflow
Manually: put the files in place
.claude/commands/agent-tool-call.workflow.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about agent-tool-call.workflow
How do I install agent-tool-call.workflow in Claude Code?
Copy .claude/commands/agent-tool-call.workflow.md into your project and use it as a command in Claude Code to generate diagrams.
What does this prompt do?
It helps you author JSON diagrams for workflows where an AI agent requests tools, receives approval or denial, executes calls to external services, and logs everything for debugging.
What diagram format does it produce?
A typed JSON specification for Archify, with your own stable block IDs and layout; the result writes to benchmark-candidate.json.
Source text
agent-tool-call.workflow.md · Apache-2.0
# Agent tool-call workflow Use the Archify skill in this repository to create a workflow diagram for an AI agent tool call. A user request reaches a chat surface and planner. A router decides whether human approval is required. Approval permits tool execution against an external provider; denial enters a blocked path that may retry. The final result returns to the user, while tool results and context are recorded for tracing. Author a fresh typed JSON diagram specification targeting the `showcase` quality profile. Choose your own stable internal IDs and layout. Make the approval decision, denied route, execution route, and external call easy to read. Use the bundled Archify CLI to validate and repair the candidate when shell access is available. The external harness will independently validate the frozen candidate. Write the final candidate to exactly `benchmark-candidate.json` in the repository root. Do not edit any other file. The candidate file, not the prose response, is the attempt 1 artifact. Do not copy a checked-in example. Do not claim that validation passed.
© tt-a1i · Apache-2.0 · tt-a1i/archify