| name | finding-review |
| description | Triages cybersecurity findings for validity, exploitability, impact, and evidence quality. Use when reviewing vulnerability reports, scanner output, pentest findings, or bug bounty submissions. |
| allowed-tools | Read Grep Glob |
Finding Review
Use this skill to assess defensive cybersecurity findings. Prioritize evidence quality and reproducibility over generic severity labels.
Workflow
- Identify the claim, affected asset, affected user role, and security boundary.
- Separate observed evidence from inference.
- Determine whether the finding is valid, false positive, duplicate, informational, or under-evidenced.
- Assess exploitability from the attacker's realistic starting permissions and network position.
- Assess impact in business and technical terms.
- Recommend the smallest remediation that removes the vulnerability class.
Output Format
Output the triage result as CSV with this exact header:
id,severity,confidence,status,category,title,location,asset_or_flow,actor,issue,impact,evidence,fix,test,scope_notes
Use status values such as valid, likely_valid, under_evidenced, duplicate, informational, likely_false_positive, or false_positive. Use category values from the finding class, such as authorization, authentication, injection, data-exposure, supply-chain, configuration, business-logic, or tests.
For reviewed findings, use:
issue: whether the original claim is supported and what security property would fail.
impact: realistic technical and business impact if valid.
evidence: observed evidence and whether it proves reachability.
fix: smallest remediation for valid or likely-valid findings.
test: safe validation or regression plan.
scope_notes: missing evidence, duplicate reference, or false-positive rationale.
Put a short explanation, missing evidence, and validation notes after the CSV unless the user asks for CSV only.
Rules
- Do not invent exploitability. If evidence is missing, say exactly what evidence is needed.
- Do not treat scanner labels as authoritative.
- Do not provide payloads, bypass chains, persistence guidance, or destructive exploitation steps.
- Prefer defensive validation steps that a system owner can run safely.
- When the target code or local test environment is available, run a safe smoke test or source-level validation before marking a finding
valid or false_positive.
- Use temporary stubs only for unavailable external services, credentials, or nondeterministic dependencies. Do not stub the claimed vulnerable control or impacted path.
- Clean up temporary fixtures, scratch files, local config changes, and ad hoc validation scripts before final output.
- When subagents are available, ask an independent subagent to challenge high-impact verdicts for evidence quality, duplicate status, exploitability, and severity.