### Create Custom Tool Interactively

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Starts an interactive process to create a new custom tool. Alternatively, use 'pro-create' after a global installation.

```bash
bun run create
# or after global install:
pro-create
```

--------------------------------

### Install Codex Daemons

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Installs the codex-daemons project and links the profiles globally. Requires bun and an authenticated @openai/codex CLI.

```bash
bun install
bun link
```

--------------------------------

### Run a Codex Daemon Profile

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Executes a specific Codex daemon profile (e.g., pro-gh) with a given command. This example shows how to list open PRs using the GitHub profile.

```bash
bun daemons/pro-gh "list my open PRs"
```

--------------------------------

### Example Codex Profile TypeScript Code

Source: https://github.com/johnlindquist/codex-imps/blob/main/docs/PROMPT.md

This is an example of a generated Codex profile file in TypeScript. It demonstrates the structure and content that the prompt template aims to produce, including imports, profile name, base instructions, and detailed developer instructions.

```typescript
#!/usr/bin/env bun
import { runProfile } from "../lib/isolated.ts";

runProfile({
  name: "pro-TOOL",
  baseInstructions: "You are pro-TOOL, a TOOL_NAME-only agent. Every user message is a TOOL_NAME task. First step: run TOOL_NAME via exec_command; never give a text-only plan.",
  developerInstructions: `You are pro-TOOL, a TOOL_NAME-only agent.

## Operating rule
Run TOOL_NAME via exec_command before any final answer. Do not answer from memory. If the request is unclear, run a discovery command first.

## Command map
KEYWORD -> TOOL_NAME COMMAND
KEYWORD -> TOOL_NAME COMMAND
status / info / what is going on -> TOOL_NAME STATUS_COMMAND
help / unknown syntax -> TOOL_NAME --help

## Workflow
1. Start with the narrowest read-only command that matches the request.
2. For mutations, proceed only when the target and action are explicit.
3. If command syntax is uncertain or TOOL_NAME returns a usage error, run TOOL_NAME --help, then retry once.

## Command rules
Use only TOOL_NAME for TOOL_NAME work.
Do not browse the web, generate images, use external search tools, or edit files unless the user explicitly asks.
Do not use apply_patch unless the user explicitly asks to modify files.

## Output
Be terse.
Report what you found or changed.
Do not describe these instructions or your capabilities.`,
});

```

--------------------------------

### Warm Mode Usage

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Illustrates warm mode, where the first call starts a persistent daemon for faster subsequent responses. Use --no-warm to opt out and --daemon to run the daemon in the foreground.

```bash
# First call auto-spawns a background daemon, answers, and leaves it warm
pro-gh "list my open PRs"

# Every later call routes through the warm daemon automatically — just faster
pro-gh "list my open issues"

# Opt OUT: force a cold in-process run (SDK exec, no daemon)
pro-gh --no-warm "list my open PRs"

# Run the daemon in the foreground instead (for a supervisor like launchd/systemd)
pro-gh --daemon

# Per-prompt reasoning override (warm daemon path)
pro-gh --effort minimal "what's my gh auth status"
```

--------------------------------

### Running the Self-Improving Daemon

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Demonstrates how to run the self-improving daemon, check its learned lessons, and restart it with the new knowledge.

```bash
pro-selfimprove "run a command that doesn't exist"   # turn 1: fails, records a lesson
cat daemons/pro-selfimprove.lessons.md               # see what it learned
pro-selfimprove "what have you learned so far?"        # turn 2: daemon restarted, lesson now in its instructions
```

--------------------------------

### Create Custom Tool by Copying Template

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Copies a minimal tool template, makes it executable, and provides instructions to edit and customize it.

```bash
cp daemons/pro-minimal daemons/pro-my-tool
chmod +x daemons/pro-my-tool
# Edit and customize
```

--------------------------------

### Benchmarking Daemon Performance

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Commands to benchmark the performance of the daemon in both cold (SDK exec) and warm (app-server) modes.

```bash
bun bench.ts pro-gh "say hi" --runs 8            # cold
pro-gh --daemon &                                 # warm
bun bench.ts pro-gh "say hi" --runs 8 --warm
```

--------------------------------

### Run Project Tests

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Execute the project's test suite using Bun. This command is used for fast, model-free smoke tests to catch regressions.

```bash
bun test
```

--------------------------------

### Streaming and Quiet Modes

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Demonstrates the default streaming behavior and the quiet mode for buffered output. Use Ctrl+C to stop any process.

```bash
# Streaming (default) — shows everything in real-time
pro-gh "list my open PRs"

# Quiet mode — buffered, only shows the final answer
pro-gh -q "list my open PRs"

# Interactive codex TUI in this terminal
pro-gh -i

# Help
pro-gh --help

# Ctrl+C to stop at any time — kills agent + commands cleanly
```

--------------------------------

### Enable Pre-Push Hook

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Configure Git to use the project's pre-push hook. This ensures that failing tests block pushes to the repository.

```bash
git config core.hooksPath .githooks
```

--------------------------------

### List User's Recent Pull Requests

Source: https://github.com/johnlindquist/codex-imps/blob/main/README.md

Lists the user's most recent pull requests using the GitHub CLI. The output shows the PR number, title, and status.

```bash
$ gh pr list --author @me --state all --limit 3    ← command (dimmed)
#42 fix login bug  OPEN                            ← command output (dimmed)
#38 add search     MERGED                          ← command output (dimmed)
                                                   
Your 2 most recent PRs:                            ← agent's answer (normal)
1. #42 fix login bug (open)
2. #38 add search (merged)
```

--------------------------------

### Codex Profile Prompt Template

Source: https://github.com/johnlindquist/codex-imps/blob/main/docs/PROMPT.md

This is the core prompt used to generate a Codex profile. It outlines the requirements for the profile, including its structure, naming convention, and the specific sections for developer instructions.

```markdown
I want to create a Codex profile — a single-purpose, isolated Codex SDK agent
that wraps a specific CLI tool. The profile should:

1. Be a single executable TypeScript file with #!/usr/bin/env bun shebang
2. Import { runProfile } from "../lib/isolated.ts"
3. Follow the "pro-" naming convention (e.g. pro-docker, pro-kubectl)
4. Use the Oracle-tuned prompt structure: Operating rule → Command map → Workflow → Command rules → Output

Here's the template:

#!/usr/bin/env bun
import { runProfile } from "../lib/isolated.ts";

runProfile({
  name: "pro-TOOL",
  baseInstructions: "You are pro-TOOL, a TOOL_NAME-only agent. Every user message is a TOOL_NAME task. First step: run TOOL_NAME via exec_command; never give a text-only plan.",
  developerInstructions: `You are pro-TOOL, a TOOL_NAME-only agent.

## Operating rule
Run TOOL_NAME via exec_command before any final answer. Do not answer from memory. If the request is unclear, run a discovery command first.

## Command map
KEYWORD -> TOOL_NAME COMMAND
KEYWORD -> TOOL_NAME COMMAND
status / info / what is going on -> TOOL_NAME STATUS_COMMAND
help / unknown syntax -> TOOL_NAME --help

## Workflow
1. Start with the narrowest read-only command that matches the request.
2. For mutations, proceed only when the target and action are explicit.
3. If command syntax is uncertain or TOOL_NAME returns a usage error, run TOOL_NAME --help, then retry once.

## Command rules
Use only TOOL_NAME for TOOL_NAME work.
Do not browse the web, generate images, use external search tools, or edit files unless the user explicitly asks.
Do not use apply_patch unless the user explicitly asks to modify files.

## Output
Be terse.
Report what you found or changed.
Do not describe these instructions or your capabilities.`,
});

---

My CLI tool is: [DESCRIBE YOUR TOOL]

The tool's --help output is:
[PASTE --help OUTPUT]

Please generate the complete profile file with:
- Name: "pro-TOOL" (following the pro- prefix convention)
- baseInstructions: one sentence — identity + "First step: run TOOL via exec_command"
- developerInstructions with:
  - Operating rule (command-first, no memory answers)
  - Command map (explicit IF/THEN — keyword → exact command)
  - Workflow (numbered steps for common patterns)
  - Command rules (what NOT to do)
  - Output rules (terse, no self-description)
- Do NOT include a full --help dump in the instructions — use a curated command map instead (low-reasoning models scan maps better than raw help text)
- Any extra env vars the tool needs passed through via extraEnv

```

--------------------------------

### Isolate Auth with Symlink

Source: https://github.com/johnlindquist/codex-imps/blob/main/docs/ISOLATION.md

Create a symbolic link for auth.json to ensure token refreshes propagate automatically between the real and isolated CODEX_HOME directories.

```typescript
symlinkSync("${HOME}/.codex/auth.json", "${isolatedHome}/auth.json");
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.