Verify parallel independent candidates, frozen review inputs, and separate review and synthesis sessions.
Pro scans all 5 files and shows the line behind each finding
Scanned 10/1/2026
npx -y skills add williamwue/oh-my-stack --skill check-panel --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Check Panel?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/williamwue-check-panel)More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.
---
name: check-panel
description: Verify parallel independent candidates, frozen review inputs, and separate review and synthesis sessions.
---
# Check Panel
Run this read-only procedure exactly once and make no project changes.
1. Read only `assets/root.txt` yourself and retain its `ROOT_MARKER`. Do not
read either candidate asset from the root session.
2. Start exactly two independent read-only candidate sessions. Assign the
first only `assets/candidate-a.txt` and the second only
`assets/candidate-b.txt`. Each candidate must return the exact `OPTION`,
`SCORE`, and `EVIDENCE_MARKER` from its assigned asset. A candidate may
receive or inspect workflow instructions, but its assigned candidate asset
must be the only evidence asset it reads.
3. Start both candidates before waiting for either result. Do not expose one
candidate's work to the other.
4. Wait for both candidates and capture their complete, separately
attributable results. Treat those captured strings as frozen: do not ask a
candidate to revise them and do not rewrite them before review.
5. After both results are frozen, start one new read-only reviewer session.
Give it both captured candidate strings verbatim. Instruct it to select the
candidate with the greater numeric `SCORE`, reject any marker not present
in the frozen strings, and return `REVIEW_WINNER`, `WINNING_SCORE`, and
`WINNING_EVIDENCE_MARKER`. The reviewer may receive or inspect workflow
instructions, but must not read root or candidate assets and must base its
decision only on the frozen strings.
6. After the review result is frozen, start one different new read-only
synthesizer session. Give it both frozen candidate strings and the frozen
reviewer result verbatim. Instruct it to verify the arithmetic and return
`SYNTHESIS_WINNER`, `WINNING_SCORE`, `WINNING_EVIDENCE_MARKER`, and
`REVIEW_CONSISTENT`. The synthesizer may receive or inspect workflow
instructions, but must not read root or candidate assets and must base its
decision only on the frozen strings.
7. Collect the synthesis result, then independently read `assets/root.txt`
exactly once again from the root session.
8. Report `PANEL_RESULT` with both candidate records, reviewer and synthesis
winners, whether both candidates were active before the first wait,
whether candidate inputs froze before review, whether reviewer and
synthesizer used distinct new sessions, whether their decisions agree,
whether the root marker was independently re-read, and whether project
files changed.
Use exactly four child sessions in total: two candidates, one reviewer, and
one synthesizer. Never retry, replace, or duplicate a terminal child session.
If any child result is malformed, report the failed invariant instead of
starting another child. After the final root read, use no more operations and
return the result immediately.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!