AI SLOP Detector — VS Code Extension v3.6.0

Real-time AI-generated code quality analysis inside VS Code. Surfaces deficit scores, structural anti-patterns, ML signals, clone detection, and actionable diagnostics without leaving your editor.

What's New in v3.6.0

CI Gate Bug Fix — `--ci-mode hard` Now Exits Non-Zero

The core CLI had a silent bug: --ci-mode hard only propagated its exit code when --ci-report was also passed. Without --ci-report, it returned 0 even on CRITICAL_DEFICIT files. This is now fixed — hard mode always gates correctly.

Pre-commit hooks were also rewritten to use python -m slop_detector.cli (the .exe wrapper on Windows did not reliably propagate exit codes):

repos:
  - repo: https://github.com/flamehaven01/AI-SLOP-Detector
    rev: v3.6.0
    hooks:
      - id: slop-detector          # hard gate — fails on CRITICAL_DEFICIT >= 70
      # - id: slop-detector-warn   # soft mode — reports only, never blocks
      # - id: slop-detector-patterns  # fast per-file pattern scan

Claude Code Skill (`/slop`, `/slop-file`, `/slop-gate`, `/slop-spar`)

Install the skill and get four agentic commands inside Claude Code:

cp -r claude-skills/slop-detector ~/.claude/skills/slop-detector

Command	What it does
`/slop`	Full project scan + interpretation + patch suggestions
`/slop-file [path]`	Per-file deep-dive with fix guidance
`/slop-gate`	CI PASS/FAIL decision with metric breakdown
`/slop-spar`	Adversarial validation — challenges "clean" findings

See docs/CLAUDE_CODE_SKILL.md for full reference.

What's New in v3.5.0

Go Language Support (Phase 3c)

The extension now activates on Go files (.go). Install the Go extra and get 6 Go-specific patterns detected in real-time:

pip install "ai-slop-detector[go]"

Pattern	Meaning
`go_empty_func`	Empty `{}` function body
`go_panic`	`panic()` crash-on-error anti-pattern
`go_fmt_print`	`fmt.Print*` debug output left in code
`go_ignored_error`	`_ = fn()` silently discarded error
`go_todo_comment`	`// TODO` inline marker
`go_god_function`	Function exceeding 50 lines

JS/TS Full Analysis (Phase 3b)

JavaScript and TypeScript files now receive full structural analysis via the dedicated JS/TS Analyzer — not just cross-language pattern matching.

Domain-Aware `--init` (Phase 3a)

slop-detector.initConfig auto-detects your project domain (library, web app, CLI tool, data pipeline) and generates a .slopconfig.yaml with domain-specific thresholds pre-tuned for that type of project.

CI Stability Fix

Fixed a jq exit 5 parse error in CI pipelines: calibration milestone hints were being written to stdout alongside the JSON payload. Hints are now correctly sent to stderr.

What's New in v3.2.1

LEDA Self-Evolving Configuration Loop

The biggest addition: the extension now exposes the LEDA calibration loop — the engine that learns your codebase's unique coding style and automatically tunes its own detection weights.

Bootstrap Config (slop-detector.initConfig): Generates a .slopconfig.yaml in your workspace root with domain-specific overrides and secure .gitignore injection. Run this once on a new project.
Self-Calibrate (slop-detector.selfCalibrate): Runs the 4D grid-search calibration (LDR × Inflation × DDC × Purity) against your analysis history, shows the result in the Output panel, and offers a one-click Apply button to write the optimal weights back to .slopconfig.yaml.

The more you use it, the smarter it gets. After 10+ analyses the LEDA engine reverse-engineers your team's past behavior and Git history to derive mathematically optimal weights. Run self-calibrate periodically — it evolves with your codebase.

Purity Dimension (4D Scoring)

The Purity score (exp(−0.5 × n_critical)) is now visible everywhere:

Diagnostics summary — appended after LDR / Inflation / DDC
Status bar tooltip — shows Purity: 0.xxx (N critical)

This completes the 4D metric display: LDR + Inflation + DDC + Purity.

Other v3.2.1 Fixes

calibrate() min_events guard: correctly handles fewer events than the minimum without raising, returns insufficient_data status
Git noise filter (P2): per-file deduplication prevents FP overcounting across runs
Black + ruff CI green across all source files

What's New in v3.1.1

Clone Detection in Core Metrics: function_clone_cluster results are now visible in the Core Metrics summary row — CRITICAL/PASS at a glance.
Workspace QuickPick: "Analyze Workspace" now shows a sorted QuickPick list of deficit files. Click any file to open it directly in the editor.
JSON parse fix: extractJson() strips [INFO] log lines before parsing.
History Trends formatting: formatted column table (Runs / Latest / Best / Worst / Trend) replaces raw JSON dump.

Features

Inline Diagnostics (Problems Panel)

Every analysis run produces structured diagnostics in the Problems panel:

Source label	Diagnostic type	Severity
SLOP Detector	Summary: score, status, LDR/ICR/DDC/Purity, ML score, Clone	Error / Warning / Info
SLOP Detector - Inflation	Unjustified jargon term at exact line	Warning
SLOP Detector - Docstring	Over-documented function (doc/impl ratio)	Error / Warning
SLOP Detector - Evidence	Unjustified quality claim lacking evidence	Warning
SLOP Detector - DDC	Unused imports summary	Info
SLOP Detector - Patterns	Structural anti-patterns (god_function, dead_code, phantom_import, function_clone_cluster, etc.)	Error / Warning / Info

Pattern diagnostics use pattern_id as the diagnostic code — VS Code's Problems panel filter works natively (e.g., filter for god_function only).

Status Bar

Right side of the status bar shows live quality at a glance:

$(check) Good (12.4)      <- score below warn threshold
$(warning) Warning (34.1) <- score >= warnThreshold
$(error) Error (67.8)     <- score >= failThreshold
$(sync~spin) Analyzing... <- analysis in progress

Tooltip includes:

Deficit Score and Status
LDR / Inflation / DDC / Purity metric values
ML slop probability (when model is present)
Clone Detection: PASS or CRITICAL

Workspace Analysis (QuickPick)

"Analyze Workspace" scans all supported files (Python, JS/TS, Go) and opens an interactive QuickPick:

$(error) src/slop_detector/cli.py          [67.8] INFLATED_SIGNAL
$(warning) src/slop_detector/core.py       [34.1] SUSPICIOUS
$(check) src/slop_detector/models.py       [11.2] CLEAN

Click any entry to open the file directly in the editor.

Lint on Save / Lint on Type

Lint on save (default: on) — triggers on every Ctrl+S
Lint on type (default: off) — triggers with 1500ms debounce

Commands (Ctrl+Shift+P → "SLOP")

Command	Description
SLOP Detector: Analyze Current File	Run analysis on active file
SLOP Detector: Analyze Workspace	Scan workspace, open QuickPick of deficit files
SLOP Detector: Auto-Fix Issues	Apply or preview (dry-run) auto-fixable patterns
SLOP Detector: Show Gate Decision (SNP)	Display SNP gate result
SLOP Detector: Run Cross-File Analysis	Detect cycles, duplicates, hotspots
SLOP Detector: Show File History	View historical score for current file
SLOP Detector: Show History Trends	Formatted table: Runs / Latest / Best / Worst / Trend
SLOP Detector: Export History to JSONL	Export full history DB to `.jsonl`
SLOP Detector: Bootstrap .slopconfig.yaml	[v3.2.1] Generate config + inject .gitignore entry
SLOP Detector: Run Self-Calibration	[v3.2.1] 4D LEDA calibration → optional Apply to config
SLOP Detector: Install Git Pre-Commit Hook	Set up pre-commit quality gate

Installation

From Marketplace

Search "AI SLOP Detector" in the VS Code Extensions panel, or:

ext install Flamehaven.vscode-slop-detector

From VSIX (Local)

code --install-extension vscode-slop-detector-3.5.0.vsix

Requirements

Python 3.9+
ai-slop-detector installed in the Python environment VS Code uses:

pip install ai-slop-detector          # core only
pip install "ai-slop-detector[ml]"    # + ML secondary signal
pip install "ai-slop-detector[js]"    # + JS/TS tree-sitter patterns
pip install "ai-slop-detector[go]"    # + Go analysis (v3.5.0)
pip install "ai-slop-detector[full]"  # everything

The extension invokes python -m slop_detector.cli <file> --json internally. Set slopDetector.pythonPath if your Python interpreter is not on PATH.

Configuration

Open Settings (Ctrl+,) and search "SLOP Detector", or edit settings.json:

{
  "slopDetector.enable": true,
  "slopDetector.lintOnSave": true,
  "slopDetector.lintOnType": false,
  "slopDetector.showInlineWarnings": true,
  "slopDetector.failThreshold": 50.0,       // deficit_score >= 50 -> Error
  "slopDetector.warnThreshold": 30.0,       // deficit_score >= 30 -> Warning
  "slopDetector.pythonPath": "python",
  "slopDetector.configPath": "",            // path to .slopconfig.yaml (optional)
  "slopDetector.recordHistory": true,       // write results to ~/.slop-detector/history.db
  "slopDetector.showCalibrationHints": true // notify when LEDA calibration milestone is reached
}

Deficit Score Thresholds

Status	deficit_score
CLEAN	< 30
SUSPICIOUS	30 – 49
INFLATED_SIGNAL	50 – 69
CRITICAL_DEFICIT	>= 70

Recommended: warnThreshold: 30, failThreshold: 50 (defaults).

Pattern Reference (v3.5.0 — 33 patterns)

Use the pattern_id as a Problems panel filter code.

Placeholder / Stub

pass_placeholder — function body is only pass
ellipsis_placeholder — function body is only ...
not_implemented — raise NotImplementedError (non-abstract)
return_none_placeholder — function body is only return None
return_constant_stub — function body is return <literal>
todo_comment — inline # TODO
fixme_comment — inline # FIXME
hack_comment — inline # HACK
xxx_comment — inline # XXX

Structural

god_function — >50 logic lines or cyclomatic complexity >10
dead_code — statements after return/raise/break/continue
deep_nesting — control-flow depth >4
empty_except — bare or typed except: pass
interface_only_class — class with ≥50% placeholder methods
nested_complexity — deep nesting AND high cyclomatic complexity combined
placeholder_variable_naming — numbered variable sequences (data1, data2…)

Quality

phantom_import — import that cannot be resolved (hallucinated dependency)
function_clone_cluster — structurally identical function bodies (AST JSD)

Cross-Language

js_var_usage — JavaScript var in Python file
js_console_log — console.log in Python file
ruby_array_each — Ruby-style .each block
go_print_format — Go fmt.Printf style

Go (pip install "ai-slop-detector[go]")

go_empty_func — empty function body {}
go_panic — panic() crash-on-error anti-pattern
go_fmt_print — fmt.Print* debug output left in code
go_ignored_error — _ = fn() silently discarded error
go_todo_comment — // TODO inline marker
go_god_function — function exceeding 50 lines

Lint Escape

noqa_overuse — excessive # noqa suppression
type_ignore_overuse — excessive # type: ignore

Development

cd vscode-extension
npm install
npm run compile    # one-time build
npm run watch      # auto-recompile on changes

Press F5 to open Extension Development Host. The Output panel channel "SLOP Detector" logs all CLI invocations and their stdout/stderr.

Changelog

See the full CHANGELOG.

v3.5.0: Go analysis (GoAnalyzer v1.0.0 — 6 patterns), JS/TS full structural analysis, domain-aware --init, CI jq fix (calibration hints → stderr).

v3.2.1: LEDA self-calibration UI (initConfig, selfCalibrate), Purity dimension in diagnostics + tooltip, min_events bugfix, black/ruff CI green.

v3.1.1: Clone Detection in Core Metrics, table style unification, Workspace QuickPick, JSON parse fix, History Trends formatting, pattern refactoring.

v3.1.0: Function clone detection (AST JSD), return_constant_stub, placeholder_variable_naming, SPAR adversarial regression, GQG scoring.

v3.0.x: Self-calibration, CI gate (--ci-mode), cross-file analysis, configurable god_function thresholds with domain overrides.

v2.9.x: Phantom import detection, history database, History Trends command.

License

MIT License — see LICENSE

AI SLOP Detector

Flamehaven