Add parallel Codex sub-agent reviews to SKILL

Update simplify-code/SKILL.md to introduce parallel review sub-agents and reorganize steps. The doc now: recommends determining the smallest diff command, uses repository local instructions to disambiguate patterns, and adds a Step 2 that launches four Codex sub-agents (Reuse, Quality, Efficiency, Clarity) with guidance on prompts, expected outputs, and recommended roles. It also clarifies aggregation of findings, discarding duplicates/conflicts, renumbers subsequent steps (fixing, validation, summary), and tightens wording about applying only high-confidence, behavior-preserving fixes and not staging/committing changes.

simplify-code/SKILL.md

@@ -1,64 +1,171 @@
 ---
 name: simplify-code
-description: "Run a multi-agent review of changed files for reuse, quality, efficiency, and clarity issues followed by automated fixes. Use when the user asks to \"simplify code\", \"review changed code\", \"check for code reuse\", \"review code quality\", \"review efficiency\", \"simplify changes\", \"clean up code\", \"refactor changes\", or \"run simplify\"."
+description: "Review a git diff or explicit file scope for reuse, code quality, efficiency, clarity, and standards issues, then optionally apply safe Codex-driven fixes. Use when the user asks to \"simplify code\", \"review changed code\", \"check for code reuse\", \"review code quality\", \"review efficiency\", \"simplify changes\", \"clean up code\", \"refactor changes\", or \"run simplify\"."
 ---
 
 # Simplify Code
 
-Review all changed files for reuse, quality, and efficiency. Fix any issues found.
+Review changed code for reuse, quality, efficiency, and clarity issues. Use Codex sub-agents to review in parallel, then optionally apply only high-confidence, behavior-preserving fixes.
 
-## Step 1: Determine the Diff Command
+## Modes
 
-Determine the appropriate diff command (e.g. `git diff`, `git diff --cached`, `git diff HEAD`) based on the current git state. If the caller specifies which diff command to use, use that. Do NOT run the diff yourself — each review agent will run it independently to keep the diff out of the main agent's context.
+Choose the mode from the user's request:
 
-If there are no git changes, review the most recently modified files that the user mentioned or that you edited earlier in this conversation.
+- `review-only`: user asks to review, audit, or check the changes
+- `safe-fixes`: user asks to simplify, clean up, or refactor the changes
+- `fix-and-validate`: same as `safe-fixes`, but also run the smallest relevant validation after edits
 
-## Step 2: Launch Four Review Agents in Parallel
+If the user does not specify, default to:
 
-Use the Agent tool to launch all four agents in a single message (`model: "opus"`, do not set `run_in_background`). Instruct each agent to run the diff command itself to obtain the diff.
+- `review-only` for "review", "audit", or "check"
+- `safe-fixes` for "simplify", "clean up", or "refactor"
 
-### Agent 1: Code Reuse Review
+## Step 1: Determine the Scope and Diff Command
 
-For each change:
+Prefer this scope order:
 
-1. **Search for existing utilities and helpers** that could replace newly written code. Look for similar patterns elsewhere in the codebase — common locations are utility directories, shared modules, and files adjacent to the changed ones.
+1. Files or paths explicitly named by the user
-2. **Flag any new function that duplicates existing functionality.** Suggest the existing function to use instead.
+2. Current git changes
-3. **Flag any inline logic that could use an existing utility** — hand-rolled string manipulation, manual path handling, custom environment checks, ad-hoc type guards, and similar patterns are common candidates.
+3. Files edited earlier in the current Codex turn
+4. Most recently modified tracked files, only if the user asked for a review but there is no diff
 
-### Agent 2: Code Quality Review
+If there is no clear scope, stop and say so briefly.
 
-Review the same changes for hacky patterns:
+When using git changes, determine the smallest correct diff command based on the repo state:
 
-1. **Redundant state**: state that duplicates existing state, cached values that could be derived, observers/effects that could be direct calls
+- unstaged work: `git diff`
-2. **Parameter sprawl**: adding new parameters to a function instead of generalizing or restructuring existing ones
+- staged work: `git diff --cached`
-3. **Copy-paste with slight variation**: near-duplicate code blocks that should be unified with a shared abstraction
+- branch or commit comparison explicitly requested by the user: use that exact diff target
-4. **Leaky abstractions**: exposing internal details that should be encapsulated, or breaking existing abstraction boundaries
+- mixed staged and unstaged work: review both
-5. **Stringly-typed code**: using raw strings where constants, enums (string unions), or branded types already exist in the codebase
-6. **Unnecessary JSX nesting**: wrapper Boxes/elements that add no layout value — check if inner component props (flexShrink, alignItems, etc.) already provide the needed behavior
 
-### Agent 3: Efficiency Review
+Do not assume `git diff HEAD` is the right default when a smaller diff is available.
 
-Review the same changes for efficiency:
+Before reviewing standards or applying fixes, read the repo's local instruction files and relevant project docs for the touched area. Prefer the closest applicable guidance, such as:
 
-1. **Unnecessary work**: redundant computations, repeated file reads, duplicate network/API calls, N+1 patterns
+- `AGENTS.md`
-2. **Missed concurrency**: independent operations run sequentially when they could run in parallel
+- repo workflow docs
-3. **Hot-path bloat**: new blocking work added to startup or per-request/per-render hot paths
+- architecture or style docs for the touched module
-4. **Unnecessary existence checks**: pre-checking file/resource existence before operating (TOCTOU anti-pattern) — operate directly and handle the error
-5. **Memory**: unbounded data structures, missing cleanup, event listener leaks
-6. **Overly broad operations**: reading entire files when only a portion is needed, loading all items when filtering for one
 
-### Agent 4: Clarity and Standards Review
+Use those instructions to distinguish real issues from intentional local patterns.
 
-Review the same changes for clarity, standards, and balance:
+## Step 2: Launch Four Review Sub-Agents in Parallel
 
-1. **Project standards**: coding conventions from CLAUDE.md not followed — import sorting, naming conventions, component patterns, error handling patterns, module style
+Use Codex sub-agents when the scope is large enough for parallel review to help. For a tiny diff or one very small file, it is acceptable to review locally instead.
-2. **Unnecessary complexity**: deep nesting, redundant abstractions, unclear variable or function names, comments that describe obvious code, nested ternary operators (prefer switch/if-else chains)
-3. **Unclear code**: choose clarity over brevity — explicit code is better than overly compact code. Consolidate related logic, but not at the cost of readability
-4. **Over-simplification**: overly clever solutions that are hard to understand, too many concerns combined into single functions or components, "fewer lines" prioritized over readability (dense one-liners, nested ternaries), helpful abstractions removed that were aiding code organization
-5. **Dead weight**: unnecessary comments, redundant code, abstractions that add indirection without value
 
-## Step 3: Fix Issues
+When spawning sub-agents:
 
-Wait for all four agents to complete. Aggregate their findings, run the diff command to get the current diff, then apply each fix directly, skipping false positives. Only edit files — do not stage, build, or test.
+- give each sub-agent the same scope
+- tell each sub-agent to inspect only its assigned review role
+- ask for concise, structured findings only
+- ask each sub-agent to report file, line or symbol, problem, recommended fix, and confidence
 
-When done, briefly summarize what was fixed (or confirm the code was already clean).
+Use four review roles.
+
+### Sub-Agent 1: Code Reuse Review
+
+Review the changes for reuse opportunities:
+
+1. Search for existing helpers, utilities, or shared abstractions that already solve the same problem.
+2. Flag duplicated functions or near-duplicate logic introduced in the change.
+3. Flag inline logic that should call an existing helper instead of re-implementing it.
+
+Recommended sub-agent role: `explorer` for broad codebase lookup, or `reviewer` if a stronger review pass is more useful than wide search.
+
+### Sub-Agent 2: Code Quality Review
+
+Review the same changes for code quality issues:
+
+1. Redundant state, cached values, or derived values stored unnecessarily
+2. Parameter sprawl caused by threading new arguments through existing call chains
+3. Copy-paste with slight variation that should become a shared abstraction
+4. Leaky abstractions or ownership violations across module boundaries
+5. Stringly-typed values where existing typed contracts, enums, or constants already exist
+
+Recommended sub-agent role: `reviewer`
+
+### Sub-Agent 3: Efficiency Review
+
+Review the same changes for efficiency issues:
+
+1. Repeated work, duplicate reads, duplicate API calls, or unnecessary recomputation
+2. Sequential work that could safely run concurrently
+3. New work added to startup, render, request, or other hot paths without clear need
+4. Pre-checks for existence when the operation itself can be attempted directly and errors handled
+5. Memory growth, missing cleanup, or listener/subscription leaks
+6. Overly broad reads or scans when the code only needs a subset
+
+Recommended sub-agent role: `reviewer`
+
+### Sub-Agent 4: Clarity and Standards Review
+
+Review the same changes for clarity, local standards, and balance:
+
+1. Violations of local project conventions or module patterns
+2. Unnecessary complexity, deep nesting, weak names, or redundant comments
+3. Overly compact or clever code that reduces readability
+4. Over-simplification that collapses separate concerns into one unclear unit
+5. Dead code, dead abstractions, or indirection without value
+
+Recommended sub-agent role: `reviewer`
+
+Only report issues that materially improve maintainability, correctness, or cost. Do not churn code just to make it look different.
+
+## Step 3: Aggregate Findings
+
+Wait for all review sub-agents to complete, then merge their findings.
+
+Normalize findings into this shape:
+
+1. File and line or nearest symbol
+2. Category: reuse, quality, efficiency, or clarity
+3. Why it is a problem
+4. Recommended fix
+5. Confidence: high, medium, or low
+
+Discard weak, duplicative, or instruction-conflicting findings before editing.
+
+## Step 4: Fix Issues Carefully
+
+In `review-only` mode, stop after reporting findings.
+
+In `safe-fixes` or `fix-and-validate` mode:
+
+- Apply only high-confidence, behavior-preserving fixes
+- Skip subjective refactors that need product or architectural judgment
+- Preserve local patterns when they are intentional or instruction-backed
+- Keep edits scoped to the reviewed files unless a small adjacent change is required to complete the fix correctly
+
+Prefer fixes like:
+
+- replacing duplicated code with an existing helper
+- removing redundant state or dead code
+- simplifying control flow without changing behavior
+- narrowing overly broad operations
+- renaming unclear locals when the scope is contained
+
+Do not stage, commit, or push changes as part of this skill.
+
+## Step 5: Validate When Required
+
+In `fix-and-validate` mode, run the smallest relevant validation for the touched scope after edits.
+
+Examples:
+
+- targeted tests for the touched module
+- typecheck or compile for the touched target
+- formatter or lint check if that is the project's real safety gate
+
+Prefer fast, scoped validation over full-suite runs unless the change breadth justifies more.
+
+If validation is skipped because the user asked not to run it, say so explicitly.
+
+## Step 6: Summarize Outcome
+
+Close with a brief result:
+
+- what was reviewed
+- what was fixed, if anything
+- what was intentionally left alone
+- whether validation ran
+
+If the code is already clean for this rubric, say that directly instead of manufacturing edits.