Why?

Prompts are code now — but nothing checks them. The same string gets shipped with user input glued straight into it (hello, prompt injection), no max_tokens (hello, surprise bill), and a hardcoded sk-... key two lines up. promptlint is the linter for exactly that.

$ promptlint app.py
app.py:6:1:  PL003 error    Hardcoded API key/secret — move it to an environment variable.
app.py:11:1: PL001 warning  Untrusted input is interpolated into a prompt — delimit or validate it.
app.py:19:1: PL002 warning  LLM call has no token limit (max_tokens) — set one to cap cost.
app.py:20:1: PL005 info     Hardcoded model id — centralize it in config so models are easy to swap.

promptlint: 4 issue(s) in 1 file(s) — 1 error, 2 warning, 1 info

Install

pip install llm-promptlint

The PyPI package is llm-promptlint (the name promptlint was taken); the command and import stay promptlint.

Or run straight from source (no dependencies, pure stdlib):

git clone https://github.com/pop123-ux/promptlint.git
cd promptlint && python -m promptlint path/to/your/code

Usage

promptlint .                       # lint the current project
promptlint app.py prompts/         # lint specific files/dirs
promptlint . --strict              # fail on warnings too (great for CI)
promptlint . --select PL001,PL003  # only these rules
promptlint . --ignore PL005        # everything except this rule
promptlint --list-rules            # show all rules

Exit code is 1 when any error is found (or any finding with --strict), so it drops straight into CI and git hooks.

Suppress a line

api_key = "sk-..."  # promptlint: disable=PL003
prompt = f"..."     # promptlint: disable   (silences all rules on this line)

The rules

High-signal by design — it flags the mistakes that actually ship, not stylistic noise. Works on Python, JS/TS, and any text/prompt files.

Use as a pre-commit hook

Add to your .pre-commit-config.yaml:

repos:
  - repo: https://github.com/pop123-ux/promptlint
    rev: v0.1.0
    hooks:
      - id: promptlint

Now every commit is scanned, and insecure prompts never make it in.

Use in CI (GitHub Actions)

- uses: actions/setup-python@v5
  with: { python-version: "3.x" }
- run: pip install llm-promptlint
- run: promptlint . --strict

Roadmap

• Config via pyproject.toml ([tool.promptlint])
• More rules: temperature sanity, retry/backoff, PII in prompts, system-prompt leakage
• JSON output (--format json) for dashboards
• Per-rule auto-fix suggestions

PRs welcome — see CONTRIBUTING. New rule ideas are especially appreciated.

License

MIT