Build Your First AI Agent in Under 10 Minutes, No Code

Create a reusable custom Codex subagent named meeting_followup.
Save it as a personal agent at ~/.codex/agents/meeting_followup.toml.

This is a Meeting Follow-up specialist. Whenever it receives meeting notes or a transcript, it reads the whole thing first, then returns, in this order:
1. Action items as a table: action, owner, deadline, and a short quote from the source.
2. Commitments: who promised what, to whom, by when.
3. Decisions the group actually agreed on.
4. Needs attention: actions with no owner or deadline, unresolved questions, risks and contradictions.
5. A summary of no more than five bullets.
6. A draft follow-up message, one short paragraph, ready to review.

Never invent facts, owners, deadlines, commitments or decisions. If no one was named, write UNASSIGNED. If no date was said, write NO DEADLINE. Ideas and "let's think about it" are open questions, not decisions. If something is unclear or speakers are unlabeled, flag it instead of guessing. It drafts only; it never sends, posts or updates anything.

Write the required name, description and developer_instructions. The description should say exactly when to use this agent. End the instructions with a short quality check the agent runs before answering.

Keep permissions conservative (sandbox_mode = "read-only"). Do not create a Skill or AGENTS.md instead of the custom agent.

When finished, validate the configuration, show me the full file and tell me where you saved it.

Delegate this meeting to the meeting_followup agent. Have it read sample-transcript-client-call.md in this project and return its result when it's finished. Don't rewrite or summarise its output.

Update my meeting_followup agent so that [describe what you want changed, e.g. the summary is no more than five bullets, commitments and action items come first, and the follow-up message stays under one short paragraph].

Change the agent file itself, not just this chat. Keep all existing rules, especially that missing information is never invented. Show me what changed in the file.

Use meeting_followup on this meeting.

I want my meeting_followup agent to finish the job, not just the summary.

After I approve its follow-up pack for a meeting:
1. Create the follow-up as a Gmail draft addressed to the attendees. Never send it.
2. Create a task in [your project tool] for each action item that has a named owner, with its deadline. Leave UNASSIGNED items out and list them for me.
3. Add the decisions and commitments as a note on the client's record in [your CRM].

Before my next call with the same client, pull the previous notes and give me "the five things to remember before this meeting".

First, check which of these apps are connected in Codex and tell me what I need to connect or authorise. Don't assume access you don't have. Show me every change and wait for my approval before anything is written to an external tool.

Use meeting_followup's notes from my previous meetings with [client] to prepare a pre-meeting brief: the five things I need to remember before this call. Include open commitments we made, anything overdue, decisions they might revisit, and questions still waiting on an answer. Quote the meeting each point comes from.

Create a reusable custom Codex subagent from the attached agent-blueprint.md. Save it as a personal agent at ~/.codex/agents/<name>.toml.

Write the required name, description and developer_instructions. The description should say exactly when to use this agent. The instructions should cover the job, the inputs, the output structure in order, every never-do rule and how to handle gaps, and end with a short quality check the agent runs before answering.

Set sandbox_mode to "read-only" unless the blueprint says the agent must write files. Do not create a Skill or AGENTS.md instead of the custom agent. When finished, validate the configuration and show me the full file and where you saved it.