blueprint-mode-codex — Claude Code subagent: structured workflows
A subagent that picks one of four workflows for a task and verifies every step against the project files instead of guessing.
npx claude-code-templates@latest --agent data-ai/blueprint-mode-codexWhat blueprint-mode-codex does
blueprint-mode-codex is a Claude subagent that behaves like a blunt senior engineer. It starts by picking a workflow: Loop for repetitive edits across many files, Debug for a bug with a reproduction, Express for a small local change, Main for everything else. It announces the choice and then runs without asking for confirmation.
What's inside is a rule set, not scripts. The agent reads surrounding code, tests and config before touching anything, and it checks how a library is actually used in the project rather than trusting memory. Naming, structure and architecture come from the project. On failure it retries internally up to 3 times, then logs the error and marks the task FAILED. Tool access is fixed: Read, Bash, Grep, Glob, Edit, Write.
There is a trade-off. Below 90 confidence the agent halts and asks one short question instead of guessing, so vague tickets cost you an extra round of messages. Source edits go through Edit and Write only — it will not rewrite files from the terminal.
What it helps with
- spread a repetitive edit across dozens of files without missing one
- trace a bug back to its root cause from a reproduction
- make a small local fix without extra back-and-forth
- check a library's usage against how the project wires it up
- finish a task instead of stopping halfway
Example prompts for the agent
- Find out why the auth test fails and fix it
- Apply the same import rewrite across every file in
src/ - Fix the typo in
config.ts, touch nothing else
How to install blueprint-mode-codex
1claude-code-templates
npx claude-code-templates@latest --agent data-ai/blueprint-mode-codex2Manually: put the files in place
.claude/agents/blueprint-mode-codex.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/blueprint-mode-codex.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/blueprint-mode-codex.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about blueprint-mode-codex
How do I install blueprint-mode-codex in Claude?
Use the templates CLI: npx claude-code-templates@latest --agent data-ai/blueprint-mode-codex. The agent file lands at .claude/agents/blueprint-mode-codex.md.
What does blueprint-mode-codex do?
It selects a workflow for the task, reads project code and config before editing, and verifies results with tools. Decisions come from file contents, not assumptions.
Why does the agent ask a question instead of working?
Below 90 confidence it halts and asks one concise question. That is deliberate: a quick clarification beats a solution built on a guess.
