Library — DistroSource

Everything digital. One source.

12 resources

Workflow · Intermediate · 45 min
Weekly agent transcript review workflow
Read a sample of an agent's real transcripts every week so drift, new failure modes and cost creep are caught before users report them.
- When: Every Friday, for any agent in production.
- 1. Sample fairly
- 2. Score each against the rubric
Steps 5Tools The agent's transcript store, A scoring sheetClaudeChatGPT
Checklist · Intermediate · 10 min
Agent tool definition review checklist
What to check before a tool is exposed to an agent, so the description and schema cannot be misread.
- Most agent misbehaviour traces to a tool the model misunderstood. Review every tool against this list before it goes live.
- ☐ The description says when to use the tool and when not
- ☐ The input schema is strict: required fields, types, enums
Items 10ClaudeChatGPT
Checklist · Advanced · 10 min
Agent production readiness checklist
What must be true before an agent is allowed to act on real systems.
- An agent is code that decides. This list is the difference between a demo and something you can leave running.
- ☐ The job and the boundary are written down
- ☐ Every side-effect tool is behind approval or a strict allow-list
Items 10ClaudeChatGPT
Code recipe · Advanced · 15 min
Tool call with schema validation and retry
Wrap a model tool call so bad arguments are caught and retried once with the error shown to the model.
- Models sometimes emit arguments that do not match a tool's schema; executing them blindly causes errors deep in your code, while failing hard wastes the run.
Language TypeScriptClaudeChatGPT
AI tool guide · Intermediate · 20 min
OpenAI Assistants and Agents for AI agents
Use OpenAI's agent tooling for tool calling, file search and structured outputs.
- OpenAI's agent APIs provide tool calling with JSON schemas, built-in file search and code execution, and structured outputs that force a response shape, which covers most single-agent business tasks without a framework.
- Workflow: Document Q&A with citations
- Workflow: Structured extraction
ClaudeChatGPT
AI tool guide · Intermediate · 20 min
Claude Agent SDK for AI agents
Build agents that use tools with the same loop Claude Code uses.
- The Agent SDK gives you the agent loop, tool definitions, permission hooks and session handling that Claude Code is built on, so an agent that reads files, runs commands or calls your APIs is configuration plus your tools rather than a hand-written loop.
- Workflow: A read-only research agent
- Workflow: A gated operations agent
ClaudeChatGPT
Automation recipe · Advanced · 45 min
Nightly agent evaluation run: automation recipe
Build nightly agent evaluation run as a GitHub Actions automation: run the agent's fixed test set every night and post the score with any regressions.
- Trigger: Schedule: nightly at 02:00 UTC.
- 1. Checkout
- 2. Run
Platform GitHub ActionsActions 4ClaudeChatGPT
Automation recipe · Intermediate · 45 min
Ticket classifier with escalation: automation recipe
Build ticket classifier with escalation as a n8n automation: classify incoming tickets and route anything uncertain or sensitive to a human queue.
- Trigger: Webhook from the help desk on new ticket.
- 1. Webhook
- 2. Anthropic or OpenAI
Platform n8nActions 4ClaudeChatGPT
Workflow · Intermediate · 45 min
Agent runbook for on-call workflow
Write the runbook an on-call person needs when an agent misbehaves at 2 a.m.
- When: Before an agent goes to production.
- 1. State the symptoms
- 2. Write the stop procedure
Steps 5Tools A document, Your monitoringClaudeChatGPT
Workflow · Advanced · 1h 30m
Human-in-the-loop approval step workflow
Add an approval gate to an agent so anything that spends money or contacts a customer waits for a person.
- When: When an agent gains a tool with external side effects.
- 1. Classify tools
- 2. Insert the gate
Steps 5Tools Your agent runtime, A queue or inbox for approvalsClaudeChatGPT
Workflow · Advanced · 40 min
Evaluate an agent with a fixed test set workflow
Run an agent against a fixed set of inputs after every change and compare against the last run, so regressions are caught before users are.
- When: After any change to the agent's prompt, tools or model.
- 1. Freeze the test set
- 2. Run and record
Steps 5Tools A test file of inputs and expected outputs, A script that runs the agentClaudeChatGPT
Workflow · Intermediate · 1h
Design an agent before building it workflow
Specify an agent on paper — job, tools, limits, failure behaviour — so the build is a transcription and not an exploration.
- When: Before writing any agent code or prompt.
- 1. Write the job and the boundary
- 2. List the tools with contracts
Steps 5Tools A document, Any assistant for critiqueClaudeChatGPT

Search