Prompts Repository

Markdown-only prompt library for Codex agents. Stores runnable prompts under prompts/ with zero-padded numbering and slugged meta-prompt artifacts under .prompts/{slug}-{purpose}/.

Contents

• prompts/create-prompt.md: create task-specific numbered prompts (NNN-name.md).
• prompts/create-meta-prompt.md: create research/plan/do meta-prompt flows saved in .prompts/{slug}-{purpose}/.
• prompts/run-prompt.md: resolve and execute prompt files (numbered or slugged) sequentially by default.
• .prompts/: slugged meta-prompt outputs with prompt.md, plan.md, research.md, etc., plus metadata blocks.
• prompts/PROMPT_RULES.md: shared baseline rules for using the helper prompts when copying them to other repos.
• AGENTS.md: repository rules and conventions.

Usage (in this repo)

• Generate a numbered prompt for a task with prompts/create-prompt.md (pick next zero-padded number). Ask your AI assistant (e.g., Codex) as shown below to do the generation.
• Create meta-prompt workflows with prompts/create-meta-prompt.md using descriptive slugs (no numbers) under .prompts/.
• Run prompts via prompts/run-prompt.md, selecting targets by number, name fragment, or explicit path (including slugged prompts).

Quick start (tell your AI)

1. “Read prompts/create-prompt.md and write the prompt to ./prompts/NNN-name.md (use the next zero-padded number) for this task: .”
2. If you need research/plan/do chaining: “Read prompts/create-meta-prompt.md and generate a slugged prompt under .prompts/{topic-slug}-{purpose}/ for this task: .”
3. To run a prompt: “Read prompts/run-prompt.md and execute <target-prompt-path> following its rules.”

Using These Helpers in Other Repos (human workflow)

• Copy the helper files in /prompts (including PROMPT_RULES.md) into the target repo (or reference them); keep helper files unnumbered and place runnable prompts as ./prompts/NNN-name.md. Do not copy any existing .prompts/ artifacts—those stay per-repo.
• Check the target repo’s AGENTS.md (or equivalent) and follow its rules; do not overwrite it. Use PROMPT_RULES.md only as the shared baseline when the target repo lacks prompt-specific guidance.
• To create a new task prompt, tell your AI assistant (e.g., Codex): “Read prompts/create-prompt.md and write the prompt to ./prompts/NNN-name.md (use the next number) for this task: ,” and let it gather repo context per the prompt.
• For research/plan/do chains, tell your AI assistant: “Read prompts/create-meta-prompt.md and generate a slugged prompt under .prompts/{topic-slug}-{purpose}/ (purpose = research/plan/do) for this task: ,” and let it gather repo context per the meta-prompt. The meta workflows include a ## Metadata block by default.
• To execute prompts, tell your AI assistant: “Read prompts/run-prompt.md and execute <target-prompt-path> following its rules.” Targets can be numbered prompts (e.g., ./prompts/003-task.md) or explicit slugged paths (e.g., ./.prompts/topic-plan/prompt.md). The runner handles sequencing and verification steps.

Conventions

• Read AGENTS.md before edits; keep prompts concise with clear sections and no placeholders; include metadata blocks for meta outputs (## Metadata with status/confidence/dependencies/open questions/assumptions).

Contributing and Testing

• No automated tests; verify changes manually (e.g., rg --files, git status --short).
• Follow AGENTS guidelines for style, naming, numbering, and sandbox awareness.