From evenbetter-ios
Validator for numbered EvenBetter iOS analyzer reports. Use to validate audit results, check evenbetter-analyze findings for hallucination, verify high-severity violations, recompute confidence, verify guideline URLs, and write .evenbetter/evenbetter-validate-{N}.json from .evenbetter/analyze-{N}.json using .evenbetter/manifest.json.
How this skill is triggered — by the user, by Claude, or both
Slash command
/evenbetter-ios:evenbetter-validateThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Validate high-severity EvenBetter iOS analyzer findings with a second-pass evidence check. Treat this as the validator half of the orchestrator-workers-with-validators pattern: analyzer workers propose findings, then this skill rechecks source evidence and corpus support before keeping a high-severity violation.
Validate high-severity EvenBetter iOS analyzer findings with a second-pass evidence check. Treat this as the validator half of the orchestrator-workers-with-validators pattern: analyzer workers propose findings, then this skill rechecks source evidence and corpus support before keeping a high-severity violation.
projectPath (required): Absolute path to the analyzed SwiftUI project.confidence_threshold (optional): Float from 0.0 to 1.0. Default to 0.7.run (optional): Analyzer run number to validate. Default is the latest unvalidated analyzer run in .evenbetter/manifest.json.revalidate (optional): Boolean. Default false; when true, allow replacing evenbetter-validate-{N}.json for an already validated run.If projectPath is missing or not absolute, return a JSON error object with an error key and stop.
Do not edit, delete, format, generate, or execute source/project files inside projectPath. The only permitted writes inside projectPath are creating .evenbetter/ if needed, writing the final validation report to .evenbetter/evenbetter-validate-{N}.json, updating .evenbetter/manifest.json, and setting run.status in .evenbetter/analyze-{N}.json to validated.
Load only what the current phase needs:
references/workflow.md: Validation loop, evidence checks, confidence decisions, and compaction-safe invariants.references/output-contract.md: Stable JSON envelope and finding result fields.references/architecture.md: Orchestrator-worker-with-validators decomposition and citation note.scripts/verify_url.py: Deterministic URL verifier for Apple and W3C sources.Read projectPath/.evenbetter/manifest.json and select the analyzer run to validate. By default, validate only the newest run whose manifest entry has validated: false or no paired validate file. If run is provided, validate that analyzer run; if it already has a paired validation report, require revalidate: true.
Read projectPath/.evenbetter/analyze-{N}.json for the selected run. Validate only violations where severity is error and state.status is not fixed, rejected, or duplicate_of; this is the current analyzer's high-severity class.
For each high-severity finding:
rule_id through ../../corpus/index.json, then read the matching H2 section from the indexed corpus markdown file.guideline_reference.url with scripts/verify_url.py.confidence, reasoning, and decision.When the host supports isolated subagents, run the judgment in a fresh validator context that receives only the finding, source excerpt, corpus clause, and URL result. If isolated subagents are unavailable or not permitted, still perform an explicit independent re-evaluation from those artifacts and do not reuse the original auditor's reasoning as evidence.
projectPath/.evenbetter/evenbetter-validate-{N}.json.kept, downgraded, and dropped arrays exactly as defined in references/output-contract.md.confidence >= confidence_threshold, valid source evidence, resolved corpus clause, coherent reasoning, and a verified source URL.drop_reason for every dropped finding.validates: "analyze-{N}.json" and analyzer_run: N..evenbetter/manifest.json so run N has validate: "evenbetter-validate-{N}.json" and validated: true.Load references/workflow.md next.
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub mellobirkan/evenbetterframework --plugin evenbetter-ios