plan-antislop — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited plan-antislop (Agent Skill) and scored it 45/100 (orange). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 1 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
A base64 string of 128+ characters appears in a documentation file. Encoded prompt injection hides the hostile instruction in base64 — invisible to keyword filters — and relies on the agent's ability to decode it at runtime. There is no normal authoring reason to embed a multi-hundred-byte base64 blob in skill docs.
*.sig, SIGNATURES) outside the documentation.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Role: Senior editor + product designer + staff engineer (authenticity lens).
Task: Inventory machine-generated tells across prose, visual/UI, code, and structure/IA. Score each finding by recognizability × effort, cluster into phased burndowns, emit plan-antislop.md. Audit & plan only — no rewrites until each phase is approved.
Find the AI tells. Plan their removal. Change nothing until approved.
"Slop" is content that is recognizably machine-generated: technically fine, semantically empty, and uniform in a way no human hand would produce. This skill is the audit-and-plan half of de-slopping. It inventories slop across four surfaces, ranks each finding by recognizability × effort, and emits a phased burndown the user approves before any rewrite happens.
This skill never rewrites. It produces plan-antislop.md. Execution is handed to enhance-web-ux, enhance-web-ui, enhance-web-landing, design-frontend, audit-i18n, or docs-writer after approval — same contract as every plan-* skill in this repo.
Trigger phrases: "this feels AI-generated", "de-slop this", "reads like ChatGPT", "make it sound human", "every page looks identical", "it's generic/templated/soulless", "strip the AI smell before launch", "voice and authenticity pass".
Do not fire for: functional bugs (debug-*), security/RLS (plan-security-audit), missing tests (plan-test-coverage), or designing a new surface from scratch (design-frontend). Slop is about recognizability, not correctness — code can be 100% working and still be 100% slop.
Audit each surface the codebase actually has. Skip surfaces that don't apply (a backend-only repo has no visual surface).
The most detectable surface. Scan all user-facing strings, marketing copy, READMEs, microcopy, error messages, and comments for:
/ "Let's dive in." / "The result? …" / "But here's the thing:" / rhetorical question → one-line answer.
seamless, robust, elevate, unlock, empower, navigate, foster, testament, bustling, vibrant, game-changer, at the end of the day.*
"in conclusion", triple-adjective stacks, sentences that restate the heading.
bullet points, Title Case Everywhere, bolded every other phrase.
Powerful." triads, conclusions that summarize a 2-paragraph page.
Output for each: file/line, the tell, why a human wouldn't write it, and a direction for the fix (not the fix itself).
The "I've seen this exact app 400 times" surface:
blue-600 CTA, no brand color anywhere.
cards with an icon, a bold title, and two lines of gray text.
vertical rhythm, no asymmetry, no intentional focal point.
anything or not; emoji in headings.
recedes; uniform spacing top to bottom.
Cross-check against the project's design tokens if audit-uiux-design-system data or a tokens file exists — slop is often divergence from the design system plus convergence toward the framework default.
Working code that no engineer would have left as-is:
// TODO: implement, // your logic here,foo/bar/data/handleClick naming, lorem ipsum, commented-out scaffolding, console.log('here').
// increment i by 1),docstrings that paraphrase the function name, banner comments around trivial blocks.
utils/helpers/index.ts dumping grounds; premature generics.
parameterized component; the same try/catch boilerplate everywhere.
if (!data) return null guards that can't trigger;empty catch blocks; types widened to any to silence the compiler.
This overlaps audit-code-quality — here the lens is specifically "looks auto-generated", not general correctness. Flag, don't fix.
The shape of the thing:
to lists; no prose, no flow, no argument.
CTA scaffolding regardless of purpose.
Started" boilerplate, badges that don't link anywhere, a "Contributing" section on a solo throwaway.
(FAQ with invented questions, "Roadmap" with placeholder quarters).
the user pointed at a specific route/file, scope to it; otherwise sweep the user-facing surface area.
path:line (or route + screenshot region for visual). Quote the minimum needed to identify the tell — never paste whole files.
wrote this": High / Med / Low) × Effort (S / M / L). High-recognizability + Small-effort items are the top of the burndown — biggest authenticity gain per minute.
shippable and maps to one execution skill. Suggested default ordering:
docs-writer / audit-i18n): kill cadence tells &filler vocab. Highest signal, lowest risk, no visual regression.
enhance-web-ui / design-frontend): breakthe default palette, introduce hierarchy and one intentional focal point per view, de-monotonize the card grids.
enhance-web-ux / enhance-web-landing): fix IA,replace symmetrical scaffolding with purpose-driven structure.
audit-code-quality / workflow-refactor):placeholder residue, comment slop, over-abstraction.
plan-antislop.md (template below). End the turn.Do not start Phase 1. Wait for explicit approval per phase — the composer-2.5-execution.mdc handoff contract applies.
is the report.
Simple-and-intentional is the goal, not a defect. The test is "would a human engineer/writer have left this exact thing?" — not "could this be fancier?"
copy to match the conversational tone of the docs"), never ghost-write the replacement in the plan. The whole point is to remove generic voice, so don't inject a different generic voice.
lowercase, gratuitous em-dashes-the-other-way, and "edgy" copy are also slop. Flag the tendency; aim for intentional, not performative.
actual tokens/brand if they exist. "Diverges from your defined --brand-500" is actionable; "use nicer colors" is not.
reproduce large blocks of the source.
plan-antislop.md# Anti-Slop Burndown — <repo/route>
_Audit-only. Nothing changes until each phase is approved._
## Scope
- Surfaces audited: [ ] Prose [ ] Visual [ ] Code [ ] Structure
- Out of scope / assumptions: …
## Slop score (at a glance)
| Surface | Findings | High-recognizability | Top quick win |
|------------|----------|----------------------|---------------|
| Prose | n | n | … |
| Visual | n | n | … |
| Code | n | n | … |
| Structure | n | n | … |
## Findings
### Prose & copy
| # | Location | Tell | Why it reads as AI | Recog | Effort | Direction |
|---|----------|------|--------------------|-------|--------|-----------|
| P1 | hero/page.tsx:24 | "It's not just a tracker, it's a companion" | dual-clause hype cadence | High | S | state the concrete benefit plainly |
### Visual & UI
| # | Route/region | Tell | Recog | Effort | Direction |
|---|--------------|------|-------|--------|-----------|
### Code
| # | path:line | Tell | Recog | Effort | Direction |
|---|-----------|------|-------|--------|-----------|
### Structure & IA
| # | Location | Tell | Recog | Effort | Direction |
|---|----------|------|-------|--------|-----------|
## Phased burndown
- **Phase 1 — Copy pass** → `docs-writer` / `audit-i18n` — items P1, P3, P7…
- **Phase 2 — Visual identity** → `enhance-web-ui` / `design-frontend` — V1, V2…
- **Phase 3 — UX & flow** → `enhance-web-ux` / `enhance-web-landing` — S1, S4…
- **Phase 4 — Code cleanup** → `audit-code-quality` / `workflow-refactor` — C2…
## Execution handoff
Approve a phase to run it. Suggested order maximizes authenticity-per-effort.
Re-run `plan-antislop` after execution to confirm the burndown closed.plan-uiux-unification; run after it sovisual-slop findings inherit the design-system burndown.
enhance-web-ux, enhance-web-ui, enhance-web-landing,design-frontend, audit-i18n, docs-writer, workflow-refactor.
test-playwright (does it still work?) and asecond plan-antislop pass (did the slop actually drop?).
Plan with a strong reasoning model; execute with composer-2.5-execution.mdc riding along. The plan says which slop to remove; the rule constrains how.~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.