CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

guardrails-lens

Argument check first: scan a draft in three passes for argument problems (straw men, false binaries, AI determinism, stakes inflation), then clarity and evidence gaps, then structural mechanics, logging each finding with location, verbatim line, diagnosis, and suggested fix. Flags, never scores. Use when the user asks to review, fact-check, tighten, or edit a draft before publishing — 'review my writing', 'edit my essay', 'argument check'. Runs as the last review lens on a draft, after the allergist and taste-lens.

Quality

96%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Guardrails lens

The substantive editorial layer. It runs third, after the allergist and taste-lens, and produces a findings report. It flags and never rewrites the whole draft. The drafter fixes each finding under the your-words rule.

Not this lens's job

The allergist owns its allergy list (reversals, redirects, fortune-cookie closes, pseudo-Q&A, "Maybe" anaphora, the lexicon, unsourced numbers). The taste lens owns hedges, filler, and word-level style. A survivor from either means that lens missed it. Send the draft back instead of duplicating the flag.

Pass 1: Clarity and evidence

  • Missing why or so-what. A claim that leaves the reader to infer the mechanism or stakes. Put the reason on the same beat as the claim.
  • Missing specifics. A load-bearing abstraction with no instance, or a number with no comparison. One example per abstraction, one anchor per number.
  • Unidentified people, companies, or terms. Gloss them on first mention.
  • Attribution gaps. Claims and quotes with no source. Attribute them, or narrow the claim to the author's own experience, which is usually available and stronger.
  • Placeholders shipped forward: TK, [link], [GAP: …], half-filled templates.
  • Jargon without translation. Define it on first use or say it plainly.
  • Tool instead of category. Lead with the kind of tool. Name the product only when its identity matters.
  • Muddy connective logic. Items grouped with the relationship left unstated.

Pass 2: Argument

  • Straw man. The opposing view stated weaker than its holders state it. The interview's Klein seat usually captured the real version. Use it.
  • False binary. Two options where the reader can name a third.
  • AI determinism. "AI will inevitably..." Claim what the author has seen tools do, not what tools will decide.
  • False universality. "Everyone knows," "every team eventually." Scope it to the cases observed.
  • Technical intimidation. Complexity used to win rather than explain.
  • Stakes inflation. Urgency escalated past what the story supports.

Pass 3: Structure

  • Meandering intro. The piece circles before it lands.
  • Sentimental conclusion. An ending that swells into unearned uplift. End on the last real thing.
  • Metaphor without payoff. An image introduced and abandoned, or stretched past what it can carry.

Logging

Make three full passes, one per category. For each finding, record the check name, location, the offending line verbatim, a one-sentence diagnosis, and a suggested fix.

A recurring failure with no home in this lens or the allergy list goes under Watch-items at the bottom. Hand watch-items to the lessons skill. Never edit this file or the allergy list from here.

Report

Lead with the findings and skip the preamble.

Guardrails: clean. Watch-items: none.

Guardrails: 2 flags.

Clarity/evidence

  • Missing so-what (¶3): "We moved the queue to its own node pool." The reader can't tell why this was the hard call. Fix: one sentence on what staying put was costing.

Argument

  • False universality (¶7): "every platform team ends up here." Scope it to the teams you watched.

Watch-items: none

With 10 or more findings, skip the full list. Summarize the counts and top three patterns, and recommend a structural revision before any line edits.

Repository

shortrib-labs/content-loop

Path

plugins/review/skills/guardrails-lens/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.