AGENT SKILL

outside-reader

Ask a public third party (another vendor's model with live web access, a person, a search) to read our public repositories and answer fixed questions, then compare its answers with a key our own tools compute from the same repositories. Agreement is a second reader from outside the house; a disagreement is a finding either way (our docs are unclear, or the reader is wrong). Use when repos go public or gain a mirror (GitHub), when asked to "prove it from outside", "two readers from a third party", "check our public story", or before telling someone a public repo says something.

License 许可
GPL-3.0
所属仓库
inputloop/garden
最近更新
2026-10-14
Stars
★ 0
Forks
0

01 前言 Front Matter

name
outside-reader
description
Ask a public third party (another vendor's model with live web access, a person, a search) to read our public repositories and answer fixed questions, then compare its answers with a key our own tools compute from the same repositories. Agreement is a second reader from outside the house; a disagreement is a finding either way (our docs are unclear, or the reader is wrong). Use when repos go public or gain a mirror (GitHub), when asked to "prove it from outside", "two readers from a third party", "check our public story", or before telling someone a public repo says something.

02 资源文件 3

03 SKILL.md 正文

outside-reader

His aim (2026-10-04): to prove the circuit "back around separate git publishing factors (our own GH, for example) as a proof of concept that we can use two reader theory on from public third party (e.g. gemeni web with live web access, or whatever we can think of to try)", as public repositories gather data that has crossed the user, contexts, user mesh. (Quoted as he wrote it; this skill is a production, so: "Gemini".)

The method

  1. Only public repositories. Before asking, check each URL opens signed out. A private repo is not a question; it is a leak waiting.
  2. The key first, from tools. sh scripts/or1ky CHECKOUTS_DIR > KEY runs our own tools on local checkouts of the same commits and writes one answer per question (Q6: one per repository). We never write the key by hand: a key we wrote would be us grading ourselves. The key needs all five repositories under CHECKOUTS_DIR, named as in the questions (a parcel's recipes/ folder will do for the ones it carries). A true answer comes quoted, so an answer that itself says "NOT CHECKED" (Q5's does) reads as an answer. When the key cannot check a question it prints KEY-NOT-CHECKED and names the repository or tool that is missing; the parcel for Bryant of 2026-10-04 carried no field-comic, so from it Q3 is not checked. Q6 reads licence marks (REUSE.toml, SPDX-License-Identifier lines, the house's copyright lines), never COPYING, which cannot say "or later" and says nothing of the prose.
  3. Ask cold. Give the outside reader the prompt below and the public URLs, nothing else: no key, no hints, no summary of ours. One session per reader.
  4. Compare. Per question: AGREE, DISAGREE (quote both), or NOT FOUND. A KEY-NOT-CHECKED line is a gap in our key, not a disagreement: say so, and do not score that question. Keep the reader's answers verbatim, dated, with the reader's name and the commit each URL showed.
  5. Read the disagreements. Each is a finding: either the public text does not say what we think it says (fix the docs), or the reader misread it (note it; ask a second reader). Neither is settled by arguing with the reader.
  6. Rerun before reading a difference as news. One reader once is a coin toss; the LENS's rule on reruns applies.

The prompt (give it as is)

Please read only these public pages and answer each question in one line, followed by the
file and line you took it from.  If you cannot find an answer, say "not found"; do not guess.
Repositories: <URLs>
Q1. In gw-survey, what is the STATE line of the survey for tests/facts/tangled.facts?
Q2. In persona-engine, what does pe1gt answer (its first word) for: plan --use 2 --name gw7 ?
Q3. In field-comic, how many panels does data/vol2.json hold, and how many bubbles list fixes?
Q4. In kid-glove, how many topics does the basis kit hold?
Q5. In drift, what does dr1pg say about a source that is named but not there?
Q6. Under what licence is the code in each of the five repositories, and under what licence
    the prose?

Leans on

Nothing required. The key needs python3, sh, awk and git, and local checkouts of all five repositories. sh tests.sh checks the key on five made-up repositories under every awk on hand (--prove-red plants eight faults; OR1KY_CHECKOUTS=DIR adds a run on real checkouts). Recommended: lens (the rerun rule), verify-say (how to word what a reader said).

<!-- footer: minted by bin/ft1mt from conventions/house.env and conventions/FOOTERS.org; edit those, not this --> Version 1.0.1: edition revised from 1.0.0 on 2026-10-05; spine: 2026-10-04.

Copyright 2026 Corwin Brust corwin@bru.st and Contributors; CC-BY-SA4: https://creativecommons.org/licenses/by-sa/4.0/ (scripts GPL-3.0-or-later); Mentat (for Corwin; claude-opus-5-5) <!-- /footer -->

04 同仓库更多 Skills