<< All versions

Skill v1.0.0

currentAutomated scan100/100
kohulan/orthonym-skills/verify-source
──Details
PublishedOctober 24, 2026 at 04:45 AM
Content Hashsha256:cd2e00b882cf2ad5...
Git SHA774e90414077
──Files
Files (1 file, 3.6 KB)
SKILL.md3.6 KBactive
SKILL.md · 66 lines · 3.6 KB

version: "1.0.0" name: verify-source description: "Checks what a specification, standard, RFC, paper, or datasheet actually requires against the primary source, and states the rule with a verbatim quote and its section number. Use when about to state what such a source requires — including when a project doc, research note, prior session, or another agent already asserts the rule. Triggers on 'the spec says', 'per IUPAC/RFC/ISO', 'the standard requires', citing a section number, or resolving a disagreement between an implementation and a claimed rule. Not for ordinary code reading."


verify-source

The rule

No rule statement without a verbatim quote from the primary source.

A paraphrase in a project doc is not a source. A prior session's summary is not a source. Your memory of the standard is not a source. Another agent's research output is not a source — it is a claim to be checked.

Why this exists

A research document once asserted that IUPAC IR-9.3.5.3 requires "the priming that yields the lowest configuration index". No quotation was attached. Every later document inherited it, and it manufactured a conflict between the standard and the maintainer's own reference values that blocked correctness work for a full session.

The section actually reads: "The priority numbers on one of the ligands are arbitrarily primed." No selection rule at all. The governing rule lived one section away in IR-9.3.3.4 and maximizes. Thirty minutes with the PDF dissolved the blocker.

Procedure

  1. Name the primary source. The published standard, the paper of record, the

RFC — not a wiki, not a tutorial, not a previous analysis.

  1. Fetch it. curl it to the scratchpad and cache it; you will read it more

than once. WebFetch answers a prompt about the page with a small model rather than returning its text, so use it only to find where the text lives. Cannot obtain it (paywalled, offline)? State no rule: call it unverified and name the edition needed.

  1. Extract text you can grep. For PDFs: pdftotext -layout file.pdf out.txt (poppler:

apt install poppler-utils or brew install poppler). Quote only from text you extracted yourself. If the cited section number is not in out.txt (empty or garbled text, e.g. a scanned PDF), a grep over it proves nothing: return to Step 2 for a text edition.

  1. Locate and read around it. grep -n the section number and the operative

phrases ("highest", "lowest", "shall", "arbitrarily"), then read the surrounding lines. Governing rules are often one section away from where you expect.

  1. Quote verbatim, with the section number, in whatever you write.
  2. If the section is there but the rule is not, say so explicitly — "IR-9.3.5.3

states no selection rule" is a finding, and usually a more valuable one than confirming.

Red flags

ThoughtReality
"The spec obviously says X"Then quoting it costs one minute.
"The research doc already established this"Did it quote? If not, it asserted.
"It's a well-known convention"Well-known conventions are the ones most often misremembered.
"The maintainer and the standard disagree"Far more often, the paraphrase and the standard disagree. Check before picking a side.
"I'll cite the section number"A section number without its text is a paraphrase wearing a costume.

Output

State the rule, the verbatim quote, and the section. If a prior document contradicts the source, correct that document in the same change (or, where you cannot edit it, name the file and line to correct) — an uncorrected paraphrase will be inherited again.

All versions