shamefile 0.1.7

A cli tool to enforce documentation for code suppressions

 

Turn silent tech debt into reviewable and documented decisions.

How it works  •  Why use it  •  Installation  •  Supported languages  •  FAQ  •  Roadmap  •  Contributing

shamefile won't let anyone silence a linter warning in your code without writing down why.

People are lazy. Both committer and code reviewer.

• The committer slaps a // NOLINT comment when there's no easy fix. They don't justify it — in most languages there's no good place for that.
• The code reviewer focuses on more important things: security, bugs, design. There's no dedicated time for checking new suppression arrivals.

Shamefile adds shamefile.yaml for the code reviewer and the shame CLI for the committer to give them tools to react before tech debt gets out of control.

How it works

Five real suppressions across five languages:

Init empty shamefile with shame me .

Fill empty why with shame next

Run again via CI with shame me . --dry-run

Why you should use it

Instantly improves AI agents

_{Ignoring warnings? Not so easy anymore. Triggers AI reflection — "is this really unfixable?". Observed across Claude, GPT, and Cursor agents. Safer vibe coding by default.}

Single source of truth

_{"Where to document?", "where to look?", "what to do?" — same answer to every question: shamefile.yaml.}

Reviewable by design

_{Make suppression review a ritual, not goodwill.}

Shamefile doesn't quit with ex-employees

_{Knowledge stays in the YAML and you can refactor with confidence.}

Installation

Source	Command
npm	npm install -g shamefile
PyPI	pip install shamefile
crates.io	cargo install shamefile
From source	cargo install --git https://github.com/BKDDFS/shamefile
Homebrew	coming soon

All channels install the shame CLI. Run shame --help to verify.

Or as a pre-commit hook:

# .pre-commit-config.yaml
- repo: https://github.com/BKDDFS/shamefile
  rev: main
  hooks:
    - id: shamefile

Commands

Command	What it does
shame me .	Scan project, sync shamefile.yaml; fail if any entry lacks why
shame me . --dry-run	CI mode — read-only validation, never writes to disk
shame next	Show first undocumented entry with the source line highlighted
shame next "<reason>"	Document the first undocumented entry inline
shame fix <location> <token> --why "<reason>"	Document a specific entry
shame remove <location> <token> (alias shame rm)	Delete a stale entry without editing the YAML by hand

Run shame --help for the full reference.

CI/CD integration

On the CI side, shame me . --dry-run is read-only and deterministic. It validates three contracts:

Check	Meaning
Coverage	Every suppression in code is registered in shamefile.yaml
Staleness	Every registered entry still points at a live suppression in code
Justification	Every entry has a non-empty why

A failure on any of the three exits non-zero.

# .github/workflows/ci.yml
- name: Check suppressions
  run: shame me . --dry-run

Flag	Description
--dry-run (-n)	Read-only validation for CI/CD — never writes to disk
--hidden	Also scan hidden files and directories (dotfiles)

Registry format

shamefile.yaml lives at the project root (git root if available, otherwise the working directory). Every entry is human-readable and stable under git diff:

---
config: {}
entries:

- location: ./src/api.py:42
  token: '# type: ignore'
  content: 'result = parse_legacy_api(raw)  # type: ignore'
  created_at: 2026-04-17
  owner: Anna Nowak <[e-mail]>
  why: 'legacy API returns untyped dict; types module in progress'

• location and token form the entry's identity.
• content is the verbatim source line — used for reconciliation when code moves.
• owner and created_at are populated automatically on first run via git blame.
• why is the only field that requires a written justification — from a developer or an AI agent. The PR reviewer decides whether the reason is good enough.

Cascade matching

A registry that breaks every time you refactor is worse than no registry. shamefile reconciles entries against source code in two passes:

1. Location match — exact file:line + token.
2. Content match — same source line + token (handles line shifts, with rename detection limited to the most recent commit via git diff HEAD~1..HEAD -M).

Reformatting a function or inserting imports above a suppression preserves the entry — owner, created_at, and why stay intact. Entries are only removed when the token itself is gone from the code.

Supported languages

shamefile is language-agnostic by design. The list below reflects what has been wired up and tested so far — not a limit on what the engine can do. If your language is missing and there's no open issue for it, please open one so we know there's interest.

Token	Tool	Language
# noqa	Flake8 / Ruff	Python
# pylint: disable	Pylint	Python
# type: ignore	Mypy	Python
# pyright: ignore	Pyright	Python
# pytype: disable	Pytype	Python
# pyre-ignore / # pyre-fixme	Pyre	Python
nosec	Bandit	Python
# pragma: no cover	Coverage.py	Python
# fmt: off / # fmt: skip	Black / Ruff	Python
# isort: skip	isort	Python
# lint-fixme / # lint-ignore	Fixit	Python
# autopep8: off	autopep8	Python
// eslint-disable, /* eslint-disable	ESLint	JS / TS / TSX
// tslint:disable, /* tslint:disable	TSLint	TS / TSX
// @ts-ignore, /* @ts-ignore	TypeScript	JS / TS / TSX
// @ts-expect-error, /* @ts-expect-error	TypeScript	JS / TS / TSX

Supported file extensions: .py, .js, .jsx, .mjs, .cjs, .ts, .tsx.

Roadmap

• MCP server — native integration for LLM-based PR authors (avoids loading the full registry into agent context)
• Custom git merge driver — auto-resolve shamefile.yaml conflicts on parallel PRs
• Additional language grammars — Rust, Go, Java, Kotlin, C# via tree-sitter
• Custom entry fields — attach ticket, reviewer, or deadline metadata to suppressions
• Native exclusion config — first-class exclude: patterns in shamefile.yaml for checked-in vendored or generated code that bypasses the default .gitignore discovery

FAQ

Why not just write the reason inline, like # noqa: F401 # legacy import?

• Reviewers don't see it. A # noqa buried in one of seven changed files rarely gets pushback. shamefile.yaml puts every suppression in the PR into one diff — the reviewer sees the full cost as a single list, with author and why per entry.
• Nothing forces a reason. Linters accept any string after the token, or none. shame me . --dry-run fails the build until every entry has a non-empty why. This matters most for AI coding agents, which lose the suppression's context the moment the session ends — the registry forces them to write the reason to disk while it still exists.
• Inline is a bad trade-off. A short reason carries no information; a useful one drowns the line of code it is attached to. The registry keeps source readable and justifications detailed.

What stops developers from writing why: 'TODO' and moving on?

The tool guarantees a string is written; the reviewer judges whether it is a real reason. If why: 'TODO' passes review, that is an organisational gap, not a tool gap — but the registry makes the gap visible: every lazy entry is one grep away, by author and date. Before shamefile, the same shortcut was hidden inside whichever file it lived in.

Won't shamefile.yaml become a merge conflict magnet on parallel PRs?

This is the same trade-off every shared-file tool (lockfiles, changelogs, schema migrations) has accepted in exchange for single-source-of-truth visibility. A custom git merge driver that resolves automatically is on the Roadmap.

The registry is sorted by (location, token), so suppressions added in unrelated parts of the codebase land in different regions of the file — most parallel PRs do not collide. When they do, shame me is idempotent: after a merge, running it on the resolved tree deterministically reconciles entries from source, so git checkout --theirs shamefile.yaml && shame me . is the escape hatch.

What about generated, vendored, or third-party code?

A repo's typical generated/vendored content is excluded for free:

• .gitignore and .ignore files are respected (handled by the same engine ripgrep uses), so node_modules/, target/, dist/, __pycache__/ etc. are skipped without configuration.
• Only .py / .js / .jsx / .mjs / .cjs / .ts / .tsx are scanned, so vendored content in any other language is silently ignored.

A first-class exclude: config in shamefile.yaml is on the Roadmap.

Contributing

Contributions are welcome. Where you start depends on what you have:

• Found a bug? Open an issue with a minimal repro.
• Idea or design question? Open a Discussion under Ideas so direction can be agreed before any code is written.
• Usage question or trouble setting things up? Ask in Q&A.
• Want to send a PR? Read CONTRIBUTING.md first — dev setup, build/test/lint commands, commit format.
• Security vulnerability? Use the private advisory form — see SECURITY.md. Do not open a public issue.

By participating you agree to the Code of Conduct.

License

shamefile is licensed under the Apache License 2.0. See the LICENSE file for more information.