Find and fix all logic bugs across your entire repository

Logic-Lens Logic Fix All is an autonomous, repository-wide audit-and-fix pipeline that runs health, review, locate/explain, fix, diff-verify, and iterate until the codebase is clean. It starts with a mandatory consent prompt (the run is token-intensive) and, once the user consents, runs hands-free through each phase: language and scope detection, scope enumeration and risk-tier classification, a health pass mapping per-module Logic Scores and L-code patterns, a deep review collecting Premises, Trace, and Divergence findings per file, conditional clarification (locating concrete failures, explaining findings whose path is unclear), a severity-sorted fix queue with a paste-ready remedy per finding, applying each fix and verifying it with a diff comparison against the original code, and iterating - re-running health and review on modified files and their consumers until the result is clean or an iteration cap is reached.

Use it when the user wants ALL logic issues in a repository found and fixed autonomously - "fix everything." For a pasted code snippet instead of a repo/directory scope, the skill skips the consent prompt and runs the fix pipeline directly on the snippet's functions. Diff verification expects a Conditionally Equivalent verdict where the differing condition is exactly the bug scenario; a fix is reverted if diff-verify instead comes back Semantically Equivalent (the fix had no effect) or shows new divergences outside the bug scenario, a regression, and is retried up to 3 times. Critical findings loop without an iteration cap; Warning and Suggestion rounds are capped by fix_all.max_iterations, with a user-escalation prompt once the cap is hit.

Input is a repository or directory scope, or a pasted code snippet, plus an optional .logic-lens.yaml with ignore, custom_risks, severity, focus, and fix_all.max_iterations settings. Output is a Fix Report containing a Scope table (role, files scanned, risk tier H/M/L, truncated flag), a Skill Invocations count per sub-skill, an Iteration History table, a Fix Log (file, lines, finding, risk, severity, fix applied, status), a Resolved-by-Clarification section for false positives found during the explain phase, an Unresolved Findings section with a reason per entry, and header fields for Logic Score before/after and counts of findings fixed by Critical/Warning/Suggestion versus unresolved.

It orchestrates several sibling skills as sub-passes: logic-health for per-module scoring, logic-review for Premises/Trace/Divergence findings, logic-locate and logic-explain for conditional clarification, and logic-diff for verifying each applied fix. It reads phase guidance from ../_shared/common.md, logic-fix-all-guide.md, and phase-specific guide files under phase-gated lazy loading, loading ../_shared/logic-risks.md, ../_shared/semiformal-guide.md, ../_shared/semiformal-checklist.md, and ../_shared/report-template.md on demand as each phase invokes them. Its report mode line is labeled "Logic Fix All".

Developers and teams who want an autonomous, evidence-verified sweep of a repository's logic bugs rather than fixing issues one at a time, and who accept a token-intensive, consent-gated run in exchange for a diff-verified Fix Report they can audit afterward.