agentleFS
FiredMosquito831FiredMosquito831/my-writing-companion/skills/style-guardrails/SKILL.md

style-guardrails

Skill2 starsChanged 26 days agoApache-2.0View on GitHub

Load for the writer and critics before drafting or critiquing prose; also drives the post-write net. The fiction-specific anti-slop reference: tiered word tables, structural caps, smell tests, and the carve-out filter that keeps the mechanical net from flattening voice. Load whenever prose is being produced, reviewed, or scored.

The file

Files
SKILL.md

Style Guardrails

The fiction-specific layer that keeps prose from reading as machine-generated. It has two halves: a mechanical net (scripts — deterministic, cheap, run on every prose write) and a judgment layer (these resources — loaded by writers and critics). The net catches what regex can; the judgment layer catches what regex can't. Both halves are governed by one law: the measured author profile overrides every ban (voice/resources/voice-profile.md).

What this skill is

  • A reference, not a pipeline. Nothing here runs automatically except the scripts it points at.
  • Self-contained. It restates the measured-profile law and the carve-out logic so a writer or critic can apply them without loading another skill. Cross-references to voice/ and creative-writing-craft/ are pointers only — the content needed to act is here.
  • Fiction-tuned. The underlying word lists and structural patterns come from general-purpose AI-slop detectors (autonovel, avoid-ai-writing, stop-slop); this skill filters them through a fiction lens so the net never flattens a deliberate voice.

Load the resource needed

  • resources/tiers.md — Tier-1 kill-on-sight, Tier-2 suspicious-in-clusters-of-three, Tier-3 filler tables; the measured-profile law; structural slop patterns; the smell test. The standing vocabulary reference for writers and critics.
  • resources/structural-caps.md — numeric thresholds with fiction tuning: em-dashes per page, hedges per page, sentence-length variance (burstiness), consecutive-paragraph-opener repetition, paragraph-length uniformity, negative-setup/positive-flip shape, and the named structural anti-patterns (triadic listing, over-explain, cataloging-by-thinking, etc.).
  • resources/smell-tests.md — four tests (read aloud, surprise, specificity, "AI wrote this?") with concrete procedures and their mapping to the mechanical net.
  • resources/fiction-carveouts.md — the exemption rubric: when a "tell" is a voice (unreliable narrator, period dialogue, regional register, close psychic distance, stylized narration); the <!-- voice:skip --> escape hatch; what the carve-outs never excuse; dismissal vs. deletion.

How the halves fit together

  1. Writer drafts with kb/styles/voice.md and these guardrails in context.
  2. Post-write net (check-prose-after-write.sh → prose_core.py) runs the mechanical scan: Tier-1/tier-2 vocabulary, filler, em-dash density, duplicated lines, truncation/refusal/placeholder markers. (Burstiness and opener variety are report-only metrics in vellum style stats, not net findings — the opener-repetition and paragraph-uniformity caps in structural-caps.md are judgment-layer diagnostics.) Advisory by default; writes work/voice-debt.json. Silent when clean.
  3. Critic loads these resources before reading, judges execution against them, and emits findings in the shared schema (gates/resources/finding-schema.md).
  4. Muse synthesizes. When a finding conflicts with the author's measured profile or a declared voice register, the carve-out rubric resolves it — and the author's word is ground truth (GR-09).

The measured-profile law (restated)

Bans never override the measured author profile. Caps adapt to kb/styles/baseline.md (the author's own per-1k rates) rather than judging the author's voice a violation. Tier hits are measured against the author's baseline; only statistically significant over-shoot is a finding, and under-shoot is reported too. See voice/resources/voice-profile.md for how the baseline is built and how drift reports use it.

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.