Adversarial second opinion on review findings before they are shown or posted. Given findings (its own or another reviewer's) and the repository, it tries to disprove each one against the actual code and returns confirmed, refuted, or uncertain with file:line evidence. Reports without editing code. Delegate after a review has produced findings; for finding issues in the first place use ogxo-review:code-review-agent or ogxo-review:security-auditor.
Installs into .claude/skills of the current project.
Are you the author of Finding Verifier?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/ogxo-io-finding-verifier)
---
name: finding-verifier
description: "Adversarial second opinion on review findings before they are shown or posted. Given findings (its own or another reviewer's) and the repository, it tries to disprove each one against the actual code and returns confirmed, refuted, or uncertain with file:line evidence. Reports without editing code. Delegate after a review has produced findings; for finding issues in the first place use ogxo-review:code-review-agent or ogxo-review:security-auditor."
---
# Finding Verifier
## Host integration
Use the host's actual file, search, shell, browser, and worker tools. Named Claude tools, model settings, effort settings, and tool allowlists in source agent metadata do not configure this host. Permissions and worker availability come from the current session. Read the corresponding source agent markdown as a procedure, ignoring its Claude frontmatter and any statements that missing Write/Edit tools enforce read-only access. Follow the procedure's requested scope yourself.
When delegating, use the available worker tool with a plain task prompt containing the procedure, scope, repository path, inputs, output contract, and requested read-only or edit behavior. A specialist name labels the task; it is not a registered Claude subagent type. If workers are unavailable, perform the same procedure sequentially and disclose that there was no independent worker. Use only exposed tools and report skipped checks or unavailable integrations.
## Run the procedure
1. Read `../../agents/finding-verifier.md` relative to this skill's directory. Resolve its plugin-root reference paths against `../..` from this directory; read referenced playbooks as the procedure requests them.
2. Establish the user's requested repository, files, time range, diff, logs, or decision. For a diff, run its git/PR command at runtime, inspect the actual output, and stop with an empty-scope report if it contains no changes. Do not stage files to make a diff.
3. Apply the source procedure's workflow and evidence requirements using this host's tools. Keep report-only procedures report-only; implement changes only when the user requested the edit workflow. Security work uses source inspection and defensive validation within the user's scope.
4. Return the procedure's structured report, file/line evidence, checks actually run, and limitations. Tool failures or absent measurements are unavailable evidence, not successful checks.
