agentleFS
peterbamuhigirepeterbamuhigire/proposal-skills/skills/meta/ai-slop-audit/SKILL.md

ai-slop-audit

Skill11 starsChanged 15 days agoMITView on GitHub

Use when auditing an existing proposal artefact for AI slop after a major iteration or before submission; use anti-ai-slop instead as the live drafting guardrail.

3 other projects use a version of this, 2 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. 3 are common, 5 are distinctive, and 3 are very specific to this project.

  • 3

    Common

    Most of the 3 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run the final A/B/C/F gate before assembly or submission.

    line 20
  2. Very specific:

    Use anti-ai-slop while creating content.

    line 24
  3. Distinctive:

    Do not grade material that cannot be inspected.

    line 25
  4. Distinctive:

    Default to read-only. Read and search are required.

    line 48
  5. Very specific:

    When a file, source, renderer, or verification capability is unavailable, mark that check not assessed and return the narrowest …

    line 52
  6. Distinctive:

    Run applicable mechanical checks, then review substance, intent, specificity, and hard cases.

    line 66
  7. Distinctive:

    Stop and grade F when any blocking marker is evidenced.

    line 67
  8. Distinctive:

    Also run as the final gate: before assembling or submitting any proposal.

    line 121
  9. Common:

    Run every applicable check.

    line 135
  10. Common:

    Every finding cites concrete evidence from the artefact: a quote, a line, a figure, a URL.

    line 194
  11. Common:

    Do not invent a flaw to pad the report.

    line 195

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI Slop Audit

Acknowledgement: Shared by Peter Bamuhigire, techguypeter.com, +256 784 464178.

The detector. Given any proposal artefact, it decides how strongly it reads as AI slop, names exactly why, and says how to fix each finding. Production-side prevention is the companion anti-ai-slop skill, and the reasoning gate is critical-analysis-business-logic.

<!-- dual-compat-start -->

Use When

  • Audit a completed proposal section, EoI, tender response, price narrative, or full bid after a major iteration.
  • Run the final A/B/C/F gate before assembly or submission.

Do Not Use When

  • Use anti-ai-slop while creating content; this skill assesses a concrete artefact in read-only mode.
  • Do not grade material that cannot be inspected; return an unassessed scope instead.

Inputs

ArtefactSource/providerRequired?Missing-input behaviour
Proposal artefact and intended audienceProposal leadYesStop and request the artefact or identify the inaccessible scope as not assessed.
Sources for factual claimsBid evidence register or verified providerConditionalMark factual verification not assessed; do not clear a blocking claim.

Outputs

ArtefactConsumerAcceptance condition
Graded slop auditAuthor and release ownerVerdict, genericness score, evidence, fix, and unassessed checks agree with the grading rules.

Evidence Produced

EvidenceConsumerAcceptance condition
Finding register and A/B/C/F verdictProposal release gateEvery finding cites a line, phrase, figure, source, or other observable artefact evidence.

Capability Contract

Default to read-only. Read and search are required; execution or network checks may verify claims when authorised. Never edit the audited artefact unless remediation is separately requested, and never submit it externally.

Degraded Mode

When a file, source, renderer, or verification capability is unavailable, mark that check not assessed and return the narrowest evidence-backed grade possible. Do not lower risk or invent findings to fill the report.

Decision Rules

EvidenceAction or gradeFailure or risk avoided
Fabricated claim, citation, client, statute, or no substantive contentGrade F and block releaseUnsupported submission
Multiple non-blocking markers or weak intentGrade C and require reworkHigh reviewer burden
Minor isolated markers onlyGrade B and list targeted fixesUnnecessary full rewrite
No blockers, low genericness, clear intentGrade A and releaseInvented audit findings

Workflow

  1. Classify the artefact and define the inspected scope.
  2. Run applicable mechanical checks, then review substance, intent, specificity, and hard cases.
  3. Stop and grade F when any blocking marker is evidenced.
  4. Recover from incomplete access by marking checks not assessed, not by assuming a pass.
  5. Issue the evidence-backed report and preserve authored material worth keeping.

Quality Standards

  • Findings are traceable, severity is consistent with the rubric, and inaccessible checks remain visible.
  • A clean artefact may receive grade A; the audit never invents debt.

Anti-Patterns

  • Reporting a vague “AI feel”. Fix: cite the exact phrase, line, figure, or formatting pattern.
  • Padding a clean audit with invented defects. Fix: allow a grade A result.
  • Editing during a read-only audit. Fix: separate diagnosis from authorised remediation.
  • Treating an inaccessible citation check as passed. Fix: mark it not assessed.
  • Removing useful authored material during cleanup. Fix: record what must be preserved.

Responsibility audit

The 25 signs are non-diagnostic editing prompts. Require exact artefact evidence before reporting a style problem, and report the reader or evaluator cost. Audit claims for scope, date, source fit, transformation, limitation, and decision effect; verify client names, experience, quotations, statistics, and examples. Separate style from fabrication, provenance, compliance, and genericness; unavailable checks remain NOT_ASSESSED.

References

  • Anti-slop production guardrail
  • Critical analysis and business logic gate
<!-- dual-compat-end -->

Machine-error audit extension

Report ME1-ME7 with exact section, page, table, or paragraph evidence. Compare adjacent sections for meaning, not just shared vocabulary. Mark NOT_ASSESSED when the Terms of Reference, experience register, or source evidence is unavailable. Do not flag required compliance repetition without checking whether it carries a distinct trace.

Impeccable-derived overlay audit

For proposal websites, decks, rendered documents, and visual bid assets, report AS1-AS7 with exact evidence and cli, browser, llm_only, or human_review mode. Purple gradients, glassmorphism, neon glow, AI-beige defaults, decorative editorial scaffolding, and decorative motion are blocking visual findings unless a task, state, accessibility, data, or approved brand reason is recorded. Mark non-visual checks not_applicable and unavailable render evidence NOT_ASSESSED.

When this runs

Cadence: run after EACH proposal section or major iteration. This is the default mode. Whenever a meaningful unit of work is finished, that is a drafted cover letter, an executive summary, an understanding-of-assignment section, a methodology, a work plan, a financial proposal, or a significant revision, run this audit on what was just produced before moving to the next section. Log the verdict. If the verdict is F (Blocked), do not progress to the next section until the blocking findings are fixed. Treat it like a quality gate that runs at every checkpoint, not a one-time review the night before the deadline.

Also auto-run on request: when the user asks to analyse, review, evaluate, audit, critique, score, or de-slop any proposal, Expression of Interest, tender response, cover letter, executive summary, methodology, work plan, team composition, financial proposal, capability statement, or business document, or asks "does this look AI-generated?" or "why does this read off?".

Also run as the final gate before assembling or submitting any proposal.

The companion anti-ai-slop skill runs continuously while content is written; this audit runs at each checkpoint to catch what slipped through.

What slop is (the yardstick)

Low-quality content produced in quantity by AI and pushed at people who did not ask for it (Merriam-Webster 2025 Word of the Year, verified). Three diagnostic properties (Kommers et al., arXiv 2601.06060): superficial competence, asymmetric effort, mass producibility. The human tell is absence of intent. In a proposal, slop is text that could be submitted to any client for any assignment. You are measuring how strongly an artefact exhibits these.

Audit method, layered, cheapest first

Step 1 — Identify the artefact and load the right checklist

Map the artefact to its proposal type: cover letter, executive summary, understanding of assignment, firm or consultant profile, relevant experience, methodology, team composition, work plan, financial proposal, or a full bid. A full proposal usually spans several. Audit each section.

Step 2 — Automated gates (machine-checkable). Any hit is hard evidence

Run every applicable check. A hit on a blocking marker (✗) fails the artefact outright.

  • 🤖 Focal-word density: delve, tapestry, realm, navigate, underscore, pivotal, intricate, leverage, foster, harness, synergy, embark, robust, vibrant, holistic, seamless, and the rest of the banned list, more than two per 500 words.
  • 🤖 Em-dash density above one per paragraph; reflexive rule-of-three; "it is not X, it is Y" repetition; uniform 15 to 25 word sentences (low burstiness).
  • 🤖 Transition clichés: "in today's fast-paced world", "let us dive in", "in conclusion", "world-class", "best-in-class", "cutting-edge" as unsupported self-praise.
  • 🤖 Mechanical formatting: Title-Case headers everywhere, excessive bold, decorative emoji, leftover tool markup ("oaicite", "contentReference").
  • ✗ 🤖 Broken or fabricated citations: dead URLs, invalid DOI or ISBN, fabricated statistics, invented client names, invented court cases or statutes, utm_source parameters copied in.
  • French: "plongeons dans", "il est important de noter que", "force est de constater", filler connectors.

Step 3 — Structural score → 0 to 100 genericness

Combine burstiness, focal-word density, duplication, and template similarity into a single genericness score. Higher means more slop-like. Report the score and its drivers.

Step 4 — Human-judgement review (the checklist no tool replaces)

  • 👁 Substance: what does this section assert or decide that required real work? If nothing, it is slop.
  • 👁 Intent and authored voice: is there a point of view, or is it relentlessly positive and free of any stance?
  • 👁 Specificity: real named clients, projects, methods, and numbers, or generic placeholders that fit any bid?
  • 👁 Hard parts: are risks, dependencies, assumptions, and the counter-case handled?
  • 👁 Proposal-specific findings (per section):
    • Cover letter and executive summary: generic value proposition that would fit any firm; inflated superlatives; no named client problem; hollow analogies.
    • Understanding of assignment: "we understand" without proof; no reference to the actual Terms of Reference, sector, or constraint.
    • Relevant experience: activities listed instead of outcomes; unverifiable or invented client references, figures, or awards.
    • Methodology and work plan: no visible logic (evidence → warrant → assumptions → counter-case → implication); frameworks used as decoration; methodology, staffing, timeline, and budget that do not tell one achievable delivery story.
    • Financial proposal: round-number filler; fees inconsistent with the staffing schedule and work plan; reimbursables that do not match the method.
  • 👁 Human-professional standard: judge proposal prose against the proposal phrase bank: needs in the client's words with a source; a cost-of-doing-nothing sentence in the understanding section; power paragraphs leading scored sections; price stated plainly next to value; a specific close. Flag standardised insight (observations any bidder could make), superlatives without adjacent proof, and statistics without a named, dated source.

Cross-check the reasoning failures with critical-analysis-business-logic, which is the complementary gate for evidence, business sense, feasibility, and achievability.

Scoring and verdict

GradeMeaningTrigger
A — CleanNo blocking hits; genericness low; substance and intent presentsubmit
B — Minor slopA few automated hits, no blockers; some genericnessfix listed items
C — SlopyMultiple automated hits or weak substance and intentrework before submission
F — BlockedAny ✗ blocker (fabricated statistic, citation, client, court case, or statute; or no substance at all)do not submit

Output format (the audit report)

# AI Slop Audit — <artefact name> — <day-month-year>
Verdict: <A/B/C/F>   Genericness score: <0–100>
Artefact type(s): <...>

## Blocking findings (✗) — must fix
- [marker] <what was found> · evidence: <quote / line / URL> · fix: <concrete action>

## Slop findings (by severity)
- [marker] <finding> · evidence: <...> · fix: <...>

## What is good (so it is not stripped in the fix)
- <substantive, specific, authored elements worth keeping>

## Recommended next step
- <rework / targeted fixes / submit>

Discipline (anti-hallucination, applies to the audit itself)

  • Every finding cites concrete evidence from the artefact: a quote, a line, a figure, a URL. No finding without evidence.
  • Do not invent a flaw to pad the report. "This section is clean" is a valid and wanted verdict.
  • Mark inferences "(inference)"; never present a guess as a measured fact.

See also

  • anti-ai-slop — the prevention companion; write proposal content so slop never appears.
  • critical-analysis-business-logic — the complementary reasoning, business-sense, feasibility, and achievability gate.
  • east-african-english and language-standards — house tone and spelling, applied on top of this skill.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.