Menu
Rewrite lazy hedges as bold narrow claims. Converges under repeated application.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
| name | sharpen |
| description | Rewrite lazy hedges as bold narrow claims. Converges under repeated application. |
| argument-hint | <file_path> |
| allowed-tools | Read, Edit, Grep |
Rewrite lazy hedges as bold narrow claims. Leave the real ones alone.
A lazy hedge is a soft word standing in for a limit the writer could have named. "A surprising fraction of X." "One important slice of X." "Arguably X." Places where the writer could have said exactly who, what, when, or how, and instead said sort of. /sharpen finds those and rewrites them. It does not touch hedges that carry real uncertainty or preserved scope.
Why the skill exists: when codex catches an overclaim, the instinctive fix is to stack qualifiers, and stacked qualifiers turn prose to mush. The right move is to narrow the claim and state the narrow version boldly, letting concrete examples do the fence-work. See feedback_narrow_and_bold.md.
Converges under repeated application. Same monoidal contract as humanize and tighten: a finite pattern set, rewrite rules that do not regenerate what they consume, a scalar norm (the hedge count from the pattern list) that tracks progress across passes. sharpen(sharpen(x)) == sharpen(x) after at most two passes.
The narrow-preserving invariant is the single rule. Every rewrite must preserve or tighten the scope of the claim. None may widen it. If no rewrite shape fits without widening, leave the hedge alone.
A little hedging is OK. The fixed point is a non-zero floor, not hedge-free prose. Real uncertainty earns a hedge. The skill's job is to clean the lazy ones, not to flatten every soft word.
If unsure, leave alone. Borderline cases are what the user is for. A successful run is "rewrote 6 obvious lazy hedges, left 9 borderline ones alone."
Soft quantifiers: "a surprising fraction of," "a lot of," "some of," "much of," "most of what X is about."
Category-introducing softeners: "one important slice of X," "one recurrent failure mode in X," "one of the failure modes of X."
Epistemic softeners: "arguably," "perhaps," "it could be said that," "in some sense," "mostly," "roughly," "somewhat" (when damping rather than expressing a degree).
Superlative softeners: "one of the most X ever" with no comparison, "extraordinarily X" with no support.
Structural throat-clearing: "it is often the case that," "it is important to note that," "it is worth mentioning that."
Not on the list (leave alone): contrastive narrowings ("not X. Y."), named fence examples ("Sweden, for example..."), scoped conditionals ("when X is Y, Z"), mathematical qualifiers ("at most α," "up to rounding"), citation hedges ("the paper argues that X"), genuine first-person uncertainty marks, and hedges inside a paragraph whose narrowing is already doing heavy lifting.
Restructure into a declarative. Fold the narrowing into a sentence whose scope is defined rather than qualified. "A surprising fraction of methodological disasters are type errors" → "When a published finding is confidently wrong rather than just noisy, the failure is almost always a type error."
Example as fence. Replace the adverb with a specific named case. Names are stronger fences than adverbs because they are verifiable. "Many governments reached for curve-fitting" → "The IHME team's early model was a curve-fitting exercise; Imperial Report 9 was not."
Narrow-then-declare. Explicit negation of the wider claim, then the bold narrow version as a separate sentence. "Most of what gets called alignment is this problem" → "AI alignment is broader than this. The failure mode that connects most directly to this essay is type-mismatch at scale."
Counts, not narratives. Hedges before, hedges after, rewrites applied, hedges left alone with one-line reasons.
Filter recent Hacker News for live threads on June's topics and hand him the list. Separately, flag the rare subset where work he has already done settles a specific quoted claim, and draft that comment as a plain argument with no link to his own writing. Use when June asks to check HN or find threads worth reading or commenting on.
Scan blog posts for AI writing tics and missed opportunities for human voice. Reports subtractions (AI patterns to remove) and additions (wordplay, arc, and claims to strengthen).
Poll June's Hacker News feeds, report what is new since the last run, and idempotently arm a background watch for replies to his comments. State persists in ~/Documents/sweep/ so it survives across sessions. Use when June asks what is new on HN, whether anyone replied, or to start watching a thread.
Argument-level compression by a strengthened absence test (does the argument stand with the same force, clarity, evidence, scope, pacing, and misreading-resistance?). Removes passages that fail it (self-referential cruft, cross-section restatement, self-recap, re-derivation, scaffolding, expired reader-orientation, stakes inflation, inert hedges, redundant examples); grafts good rhetoric onto an adjacent essential point rather than deleting. Idempotent. The structural cousin of /tighten (word-level) and /not-but (negation).
Audit a technical candidate's resume/CV by verifying its claims against the public contribution graph instead of taking the text at face value. Sorts every claim onto the self/peer/world attestation lattice, applies a cost-and-stake filter so cheap proxies (stars, downloads, backlinks) don't count, runs LIVE checks (GitHub, arXiv, package registries, CVEs), and reports two separate axes — merit (staked acceptance of the candidate's specific work) and attention (timestamped output) — plus a verifiability score in [0,1]. For technical recruiters and hiring managers. Invoke when given a candidate — a GitHub handle, a resume/CV, an arXiv author, a personal site — and asked to verify, audit, vet, or check their claims. Runs at two depths — a quick triage scan or a deep per-claim audit — and elicits which. Checks the provenance of a personal site or self-hosted resume (Internet Archive + git history) as an un-backdatable anti-tailoring anchor. Needs tool access (Bash/curl or WebFetch).
Find and defuse "not X but Y" / "isn't X; it's Y" / "X — not Y" constructions. Deterministic grep for detection (no LLM blind spot), the not-but subagent for triage. Split-or-cut tail negations ("X, not Y") aggressively — promote the point to its own sentence or delete it; recast in place only mid-clause "not X but Y" that genuinely can't be split; keep just the rare iconic thesis line literal.
