A multi-language codebase health scanner that combines mechanical detection (dead code, duplication, complexity, security) with LLM-based subjective review (naming, abstractions, module boundaries), then works through a prioritized fix loop. This Open Paws fork adds 65 advocacy language rules, a 3-adversary activist security detector, and persona-based browser QA — all integrated into the same scoring and queue system as the upstream detectors.
Note
This project is part of the Open Paws ecosystem — AI infrastructure for the animal liberation movement. Explore the full platform →
Requires Python 3.11+.
# Install from this fork
pip install "git+https://github.com/Open-Paws/desloppify.git#egg=desloppify[full]"
# Exclude generated directories, then scan
desloppify exclude node_modules dist
desloppify scan --path .
# Run the fix loop: next → fix → resolve → repeat
desloppify nextOr without installing:
uvx --from "git+https://github.com/Open-Paws/desloppify.git" desloppify scan --path .Add .desloppify/ to .gitignore — it holds local state that should not be committed.
Overall score = 25% mechanical + 75% subjective. A score above 98 correlates with a codebase a seasoned engineer would call well-crafted. The scoring resists gaming — the only path upward is genuinely better code.
Mechanical pool (25% of overall)
| Dimension | Weight |
|---|---|
| File health | 2.0 |
| Code quality | 1.0 |
| Duplication | 1.0 |
| Test health | 1.0 |
| Security | 1.0 |
| Advocacy language (fork addition) | 1.0 |
| Advocacy security (fork addition) | 1.0 |
| Persona QA (fork addition) | 1.0 |
Subjective pool (75% of overall) — scored by an LLM against a blind review packet. Includes high/mid/low elegance, contracts, type safety, design coherence, abstraction fit, logic clarity, naming quality, and more.
Minimum score thresholds (Open Paws internal policy):
- Gary (autonomous agent): ≥ 80
- Platform repos: ≥ 75
- All other repos: ≥ 70
29 languages. Full plugin depth for TypeScript, Python, C#, C++, Dart, GDScript, Go, and Rust. Generic linter + tree-sitter support for Ruby, Java, Kotlin, and 18 more.
Advocacy language detector — 65 rules
Detects speciesist language in code, comments, and documentation across all 29 languages plus .md, .txt, .rst. Rules sourced from no-animal-violence.
| Category | Count | Examples |
|---|---|---|
| Idioms | 30 | "kill two birds with one stone", "beat a dead horse" |
| Metaphors | 21 | "sacred cow", "cash cow", "sacrificial lamb" |
| Insults | 6 | "code monkey", "cowboy coding" |
| Process language | 5 | "nuke", "cull", "kill process" |
| Terminology | 3 | "master/slave", "whitelist/blacklist" |
Each finding includes a suggested replacement. Context suppression reduces false positives for technical terms (POSIX kill(), git master branch), proper nouns, and quotations.
Advocacy security detector — 3-adversary threat model
Heuristic detection for activist protection antipatterns against three adversaries:
- State surveillance — ag-gag statutes, warrants, device seizure
- Industry infiltration — corporate investigators, social engineering
- AI model bias — training data encoding speciesist defaults, telemetry leakage
Detects: identity leakage in logs/errors, sensitive data routed to external AI APIs without zero-retention headers, investigation materials in public paths, unencrypted writes of sensitive data, IP address logging, sensitive data in browser storage.
Persona-based browser QA
desloppify persona-qa --prepare --url https://example.com # generate agent instructions
# agent runs browser testing and captures findings in JSON
desloppify persona-qa --import findings.json # merge into state
desloppify persona-qa --status # per-persona summary
desloppify next # persona QA items appear in queueWindows platform fixes
input()blocking in TypeScript logs detector replaced withisatty()guardmsvcrt.locking()infinite wait replaced with 5-second retry timeout- Dataclass JSON serialization crash on state save fixed with
dataclasses.asdict()fallback
desloppify next # get the top-priority item; it shows which file and the resolve command
# fix the code
desloppify plan resolve # mark it done
desloppify next # get the next itemState persists across sessions in .desloppify/ so the tool chips away over multiple runs.
| Document | Purpose |
|---|---|
| docs/SKILL.md | Full agent workflow guide (scan → review → plan → execute) |
| docs/CLAUDE.md | Claude Code integration |
| docs/CURSOR.md | Cursor integration |
| docs/CODEX.md | OpenAI Codex integration |
| docs/GEMINI.md | Gemini integration |
| docs/WINDSURF.md | Windsurf integration |
| docs/QUEUE_LIFECYCLE.md | Queue and plan lifecycle |
| desloppify-fork-architecture.md | Fork architecture and extension points |
| persona-qa-architecture.md | Persona QA design |
Install agent skill files directly from the CLI:
desloppify update-skill claude # options: claude, cursor, codex, copilot, droid, windsurf, geminiInternal structure
The package is organized into five main layers:
desloppify/languages/ — per-language configs. Each language is a LangConfig dataclass with ordered DetectorPhase callables, dependency graph builder, function extractor, and subjective review dimensions. Full plugin depth: TypeScript, Python, C#, C++, Dart, GDScript, Go, Rust. Generic support via tree-sitter for 21 more languages.
desloppify/engine/ — scoring, plan, work queue, and scan workflow. Scoring uses two pools (mechanical 25%, subjective 75%) with per-dimension weighted averages and sample dampening for small codebases. Plan and queue state persist in .desloppify/ with file locking for safe concurrent access.
desloppify/intelligence/ — subjective review logic. Prepares blind review packets and normalizes LLM assessment output into the same Issue format as mechanical detectors. Never calls an LLM directly — model selection happens at the orchestration layer, keeping this layer model-agnostic.
desloppify/app/ — CLI commands and orchestration. Each command (scan, next, review, persona-qa, etc.) lives in its own subdirectory. The command registry is a plain dict mapping.
desloppify/base/ — detector metadata catalog (65+ built-in entries), scoring policy registry, and the DetectorMeta / LangConfig type contracts.
Fork-specific files:
desloppify/engine/detectors/advocacy_language.py— shells out to semgrep/eslint/vale for the 65-rule advocacy language checkdesloppify/engine/detectors/advocacy_security.py— heuristic activist security checksdesloppify/app/commands/persona_qa/— browser QA command and persona profile handling
Upstream tracking:
This fork tracks peteromallet/desloppify as upstream. Fork-specific changes live in new files and minimal patches to scoring constants and language configs. Upstream merges are designed to stay clean.
git fetch upstream
git merge upstream/main
Issues and pull requests go to github.com/Open-Paws/desloppify.
For bugs in the upstream scanner unrelated to the advocacy extensions, file at github.com/peteromallet/desloppify.
Contributions to the advocacy language rules (new patterns, replacements, context suppressions) are especially welcome — the rule set lives in Open-Paws/no-animal-violence and is shared across the full tooling suite (semgrep, ESLint, Vale, pre-commit, GitHub Actions).
If you maintain an open-source project and want to add compassionate language checks, see project-compassionate-code for the broader initiative.
MIT — same as upstream. See LICENSE.
Upstream: peteromallet/desloppify by Peter O'Malley.
Advocacy rules: Open-Paws/no-animal-violence.
Donate · Discord · openpaws.ai · Volunteer