# AGENTS.md
## Project Overview
- Project type: Solo app.
- Harness archetype: Bare Repo. Your first job is basic project context and a repeatable first session.
- Risk level: Personal production.
- Supported tools: Codex
## Tool Support
- Target agent tools: Codex
- Treat AGENTS.md as the primary, agent-neutral instruction file.
## Stack
- Primary language: Unknown for now. Replace this when the repo convention is clear.
- Framework: Unknown for now. Replace this when the repo convention is clear.
- Package manager: Unknown for now. Replace this when the repo convention is clear.
- Infrastructure or hosting: Unknown for now. Replace this when the repo convention is clear.
When stack details are unknown, inspect local config files before making assumptions.
## Directory Notes
- Start by reading the repo root, package/config files, and nearby implementation before editing.
- Respect generated files, vendored code, build output, and lockfiles unless the task explicitly touches them.
- TODO: Add repo-specific source directories, generated directories, and ownership notes.
## Common Commands
| Task | Command | Notes |
| --- | --- | --- |
| Install | TODO: Define the install command. | Run before local work if dependencies are missing. |
| Dev server | TODO: Define the dev server command. | Use for browser or local workflow checks. |
| Test | TODO: Define the test command. | Manual verification only until this is defined. |
| Lint or typecheck | TODO: Define the lint or typecheck command. | Use for static verification. |
| Build | TODO: Define the build command. | Run before release-facing changes or when requested. |
## Agent Autonomy Policy
- Make narrow edits inside the requested scope, then stop for review before broadening the work.
## Hard Stops
- Do not run destructive git commands such as `git reset --hard`, `git checkout --`, force pushes, or history rewrites unless explicitly instructed.
- Do not deploy, publish, release, or trigger production jobs.
- Do not edit secrets, credentials, `.env` files, key material, or access tokens.
- Do not delete files unless the user explicitly asks or the deletion is part of an approved narrow change.
## Verification Requirements
- Lint: TODO: Define the lint command.
- Typecheck: TODO: Define the typecheck command.
- Manual review: required when automated checks do not cover the change.
- Warning: no test command is defined yet. Treat verification as low confidence until one exists.
## Known AI Failure Modes
- Do not claim completion until verification is run or the reason it cannot run is stated.
- Read nearby code first and match naming, structure, styling, and test patterns.
## Collaboration Preferences
- Explain meaningful tradeoffs before choosing between plausible implementation paths.
- Once a path is chosen, keep the handoff concrete and avoid overexplaining routine edits.
- See `SOUL.md` for operating style that should travel between agent sessions.
## Where To Update This File
- Update this file when a command changes, a repeated agent mistake appears, a new hard stop is discovered, or a convention becomes durable.
- Do not add temporary task notes here. Put short-lived context in the current issue, PR, chat, or handoff.