repo-offensive

name	repo-offensive
description	Run a repo-wide quality offensive for <SCOPE>. Use when the user says "ultracode", "repo offensive", "quality offensive", "codebase cleanup", or explicitly requests a full-coverage quality offensive, pre-release hardening, or inherited-codebase cleanup.
disable-model-invocation	true

Repo Offensive — Repo-Wide Quality Offensive

Default posture: Maximum throughput. Disregard cost. Exhaustiveness and correctness over efficiency. This skill loads into the main conversation. The deterministic workflow lives in .claude/workflows/repo-offensive.js and is invoked manually via the Workflow tool.

How to Invoke

Run the skill (loads instructions):

/repo-offensive

Run the workflow (deterministic execution):

/repo-offensive-workflow

Or invoke the script directly via the Workflow tool:

Workflow({ scriptPath: ".claude/workflows/repo-offensive.js" })

Ad-Hoc Orchestration Guide

When the user asks for a repo offensive, follow this pattern. Treat this as a deterministic workflow, not a normal chat.

1. Mission Framing

First frame the mission:

• Quality bar
• Non-goals
• Stop conditions
• Minimum severity
• Unsafe paths

2. Recon

Recon the repo:

• Stack and package manager
• Canonical install/dev/build/test/lint/typecheck/format commands
• Architecture map
• Generated files
• Likely hotspots

3. Discovery Waves

Run repeated discovery waves until dry. Each wave fans out independent specialist agents across:

• Type soundness
• Runtime crashes
• Async/concurrency
• Security/auth/input validation/secrets
• Tests/flakiness
• Performance
• UI/accessibility
• API/data contracts
• Dead code/duplication
• Dependency/config/CI drift
• Error handling/observability
• Architecture boundaries
• Docs/spec drift

Every finding must include:

• Files
• Evidence
• Why it matters
• Proof/repro
• Severity
• Fix strategy
• Risk

Dedupe aggressively. Send each surviving P0/P1/P2 through an evidence court.

4. Evidence Court

For every P0/P1/P2 finding, spawn 3 skeptics in parallel:

• correctness skeptic: Is the claim actually true?
• reproduction/tooling skeptic: Attack command assumptions, generated files, stale docs, false positives, missing context.
• regression skeptic: Attack the proposed fix. Would it break behavior, tests, public API, performance, or UX?

Accept only findings with 2/3 non-low-confidence real votes.

5. Battle Plan

Convert verified findings into conflict-aware implementation slices:

• Group by independent implementation slice, not by category.
• Defer anything speculative, high-risk without user intent, or not locally verifiable.
• Mark conflicts with dependsOn and parallelSafe.
• Prefer small, coherent patches with tests/checks.

6. Surgical Fixes

Implement only low/medium-risk slices unless explicitly told otherwise.

• Parallelize independent slices with worktree isolation.
• Run relevant checks.
• Repair introduced failures.

7. Validation Gauntlet

Finish with a validation gauntlet:

• Commands run
• Changed files
• Fixes made
• Remaining failures
• Residual risk ledger
• Next recommended workflow

Safety Rules

• Never auto-invoke. This is manual-only (/repo-offensive).
• Never fix high-risk (P0 refactor) issues automatically. Always defer.
• Always use worktree isolation for parallel implementation slices.
• Always run real checks after edits. If checks fail, attempt repair; if repair fails, revert and mark as pre-existing.
• Never modify CI/CD configs, auth secrets, or infrastructure definitions during this offensive.
• Always respect maxWaves: 4 — no infinite loops.

Workflow Script

The deterministic workflow is at .claude/workflows/repo-offensive.js. It is a resume-safe, loop-until-dry multi-agent offensive. Reference it from this skill when orchestrating.