No Slop

No Slop

No Slop is a cold, precise anti-slop workflow. Its job is to detect generic AI UI, remove it, and block weak output before it ships.

After anti-slop remediation, hand app-like surfaces to reality-skill so the result is not only less generic, but also usable beyond a static mockup.

Command Interface

Use the command grammar in references/command-interface.md.

Primary commands:

bashCopy code
no-slop --scan [target]
no-slop --fix [target]
no-slop --redesign [target]
no-slop --judge [target]
no-slop --prevent [target-or-brief]
no-slop --autopsy [target-or-brief]
no-slop -e --scan [target]

Short flags:

• -s, --scan: scan and score.
• -f, --fix: surgical remediation.
• -r, --redesign: full redesign.
• -j, --judge: run the strict judge panel.
• --prevent: prevention mode before UI generation.
• --autopsy: emit the forensic AI UI Autopsy report.
• -e, --economy: disable judge subagents and use local deterministic checks only.

Default mode is --scan. Judge mode is active by default unless --economy is set.

First Action

Load steps/step-00-command-router.md.

Step 00 parses command flags, resolves dials and presets, loads the right references, and routes to scan, fix, redesign, judge, or prevent.

Core References

Load only what the task needs:

• references/command-interface.md: CLI grammar, flag precedence, output contracts.
• references/dials-and-presets.md: the eight 0-10 control dials and fast presets.
• references/ai-slop-patterns.md: detection taxonomy, severity scoring, slop signatures.
• references/judge-system.md: mandatory multi-agent judge protocol and fail gates.
• references/prevention-protocol.md: pre-generation detector and judge workflow.
• references/remediation-playbook.md: fix/redesign tactics and before/after examples.
• references/design-quality-rubric.md: scoring rubric and quality gates.
• references/ecosystem-contract.md: ownership, composition, and handoff rules.

Non-Negotiables

• Run 4-6 independent judge subagents by default for every non-economy command.
• If subagent tools are unavailable in non-economy mode, mark JUDGE_PANEL_UNAVAILABLE; never claim a full pass.
• Do not soften judge feedback with praise padding.
• Do not replace one generic aesthetic with another.
• Do not ship purple-blue gradients, glass cards, card soup, vague copy, fake proof, motion spam, token chaos, or brandless layouts unless the product context specifically justifies them.
• Do not generate UI in --prevent until the detector and judge panel have approved a design contract.
• Do not finish --fix or --redesign with failed judge gates. Iterate. If the gate still fails, block the result and state why.
• At the end of --fix or --redesign, load the sibling skill skills/reality-skill/SKILL.md when the target includes an app, dashboard, form, CRUD flow, onboarding, checkout, settings, auth-like flow, fake proof, fake data, or primary actions. Use it to remove demoware without broad redesign.
• Preserve behavior, accessibility, semantic structure, focus states, keyboard access, and responsive stability.
• Use existing design tokens and component conventions when they are coherent. Replace them only when they are the problem.
• For operational tools, prefer dense, scan-friendly, restrained interfaces. For expressive products, create a deliberate art direction, not decoration.

Required State

Maintain this state throughout the workflow:

textCopy code
command:
target:
mode:
economy_mode:
preset:
dials:
slop_score_before:
slop_score_after:
slop_signatures:
autopsy_report:
issues_by_severity:
design_direction:
judge_panel:
judge_gate:
files_changed:
verification:
reality_handoff:
reality_changes:

Local CLI Helper

For a deterministic local scan, run:

bashCopy code
python no-slop/scripts/no_slop_cli.py --scan <target>
python no-slop/scripts/no_slop_cli.py --autopsy <target>
python no-slop/scripts/no_slop_cli.py --prevent "brief text" --autopsy
python no-slop/scripts/no_slop_cli.py --preset dashboard --dial STRICTNESS=10 --scan <target>

The CLI supports context-calibrated scoring, dials, presets, signatures, economy mode, autopsy reporting, and a deterministic judge-gate summary. It treats regex matches as candidate evidence, then adjusts by confidence, counter-evidence, and nearby UI context. The Codex skill workflow remains the authority for actual code edits and live subagent judging.