By Emasoft
Ultracode code-audit engine for Claude Code: one deterministic map-filter-reduce workflow (scripts/workflows/caa-engine.js) drives every operation over swarms of opus agents — pre-commit gate of staged files, whole-codebase or scoped scan, delta recheck since a git ref, scan-and-fix with per-fix adversarial verification, and GitHub PR review with claim-verification and cross-layer lenses plus 21 distilled domain lenses (docker, solidity, iOS, GraphQL, frontend, JWT, prompt-injection, ...). Every finding is adversarially verified before it is reported; one consolidated report plus machine-readable findings.json per run.
Based on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Legacy alias for the ultracode codebase audit. Superseded by /caa-scan (scan-only) and /caa-scan-and-fix (audit + root-cause fixes). Kept so existing "audit the codebase" muscle memory still resolves; it simply redirects to the new ultracode commands.
Legacy alias for the incremental (changed-since-a-ref) ultracode audit. Superseded by /caa-delta. Kept so existing "audit the changes" / "incremental audit" muscle memory still resolves; it simply redirects to the new ultracode command.
Recheck recent changes — scan-only audit of files changed since a git ref (default origin/main…HEAD), via the shared caa-engine (map → filter → reduce). Optionally traces direct dependents of the changed files. No fixes, no git writes. Final report → reports/code-auditor-agent/.
Compare multiple candidate implementations of the same task against ONE fixed input/contract via the shared caa-engine (task=impl-compare): map (one opus evaluator per candidate, scoring correctness/edge-cases/performance/quality) → filter (adversarial verify of each correctness verdict) → reduce (a ranking matrix naming the winner). The fixed input is cache-shared across every candidate; only the candidate script varies. No edits, no git writes. Final report → reports/code-auditor-agent/.
PR review — ultracode audit of a GitHub PR via the shared caa-engine `pr` lens-set: per-file scan of the changed files PLUS two PR-unique once-per-run lenses (claim-verification: PR description vs actual diff; cross-layer: cross-file mismatches) → one PR-review comment with a PASS/CONDITIONAL/FAIL verdict. Scan-only (no fixes). Final report → reports/code-auditor-agent/.
Trigger with /audit-codebase, 'audit the codebase', 'compliance audit', 'codebase audit', 'scan and fix the code'. Use when auditing a codebase (whole, scoped, or changed-since-a-ref) and optionally applying fixes.
Trigger with 'review and fix the PR', 'review the PR and apply fixes', 'pre-merge review and fix'. Use when reviewing a GitHub PR AND then applying root-cause fixes to its changed files.
Trigger with 'review the PR', 'check the PR', 'audit the PR', 'pre-merge review'. Use when reviewing a GitHub PR or running a pre-merge quality gate.
Trigger with "/caa-extended-audit", "generate scenarios", "discover entry points". Use when an extended audit needs end-to-end scenario walks beyond line review. Emits scenarios.json for the ultracode engine's scenario-walk lens.
Trigger with "ecaa self test", "run efficacy gate". Use when verifying the ultracode engine catches every seeded bug in the bundled fixtures — before release or on a schedule. Runs the pytest gate plus one engine pass.
Uses power tools
Uses Bash, Write, or Edit tools
Version: 4.1.1 License: MIT Author: Emasoft
Ultracode code-review engine for Claude Code. ONE deterministic map → filter → reduce
workflow (scripts/workflows/caa-engine.js, run by the Workflow tool) drives every
operation over swarms of opus agents: pre-commit gating, whole-codebase or scoped scans,
delta rechecks, scan-and-fix with per-fix adversarial verification, and GitHub PR review
with claim-verification, cross-layer, and skeptical whole-diff lenses — plus 21 distilled
domain lenses (docker, solidity, iOS, GraphQL, elixir, frontend, monorepo, i18n, l10n,
JWT, prompt-injection, logging, MCP-server, API design, type design, assumptions,
function contracts, pre-mortem, architecture consistency, scenario-walk, skeptical).
Every finding is adversarially verified by a second, independent reviewer before it is
reported — refuted and downgraded findings are listed with the evidence that killed
them, never silently dropped.
Requirements: Claude Code v2.1.154 or later (the Workflow tool) with uv/uvx and opus
access at session effort xhigh or max for the ultracode path. Without the Workflow tool
(ultracode disabled in settings/env), or with CAA_ULTRACODE=0, the commands fall back to a
simple inline scan at any effort — same reports, lower fidelity (no agent swarm). See
Ultracode vs. simple-scan fallback.
Install from the emasoft-plugins marketplace:
/plugin install code-auditor-agent@emasoft-plugins
After installing, run /reload-plugins to activate without restarting.
| Scope | Command | Use case |
|---|---|---|
| User (default) | /plugin install code-auditor-agent@emasoft-plugins | Personal use across all projects |
| Project | claude plugin install code-auditor-agent@emasoft-plugins --scope project | Shared with team via .claude/settings.json |
| Local | claude plugin install code-auditor-agent@emasoft-plugins --scope local | Project-specific, gitignored |
For local development, launch Claude Code with the plugin directory:
claude --plugin-dir /path/to/code-auditor-agent
Add the marketplace to your project's .claude/settings.json so team members get
prompted to install it automatically:
{
"extraKnownMarketplaces": {
"emasoft-plugins": {
"source": {
"source": "github",
"repo": "Emasoft/emasoft-plugins"
}
}
},
"enabledPlugins": {
"code-auditor-agent@emasoft-plugins": true
}
}
Raise the session effort first — the engine is opus-only and every command halts below
xhigh:
/effort max
Then pick the command that matches the job:
| You want to… | Command |
|---|---|
| Gate the STAGED files before a commit (PASS/FAIL verdict) | /caa-precommit |
| Audit the whole repo, or an explicit path/glob, scan-only | /caa-scan [paths...] |
| Recheck only the files changed since a git ref (+ dependents) | /caa-delta [ref] [deps] |
| Audit AND apply root-cause fixes (in place, fix-verified) | /caa-scan-and-fix [paths...] |
| Review a GitHub PR (ready-to-post review comment) | /caa-pr-review <pr-number> |
| Audit a codebase against a spec / requirements doc (MISSING + VIOLATING) | /caa-spec-audit <spec> [paths...] |
| Compare candidate implementations of one task against a fixed input | /caa-impl-compare <input> <impl...> |
Examples:
/caa-precommit
/caa-scan scripts/ skills/ min-severity=MAJOR
/caa-scan src/chat component=chat-audit lenses=docs/invariants.md
/caa-delta origin/main deps
/caa-scan-and-fix scripts/foo.py
/caa-pr-review 206
/caa-spec-audit design/requirements/PRRD.md src/
/caa-impl-compare bench/contract.md impls/v1.py impls/v2.py
Shared knobs (all commands): conc=N concurrent auditors (default 6, max 16);
component=NAME sub-folder for the reports; min-severity=SEV body filter;
lenses=p1,p2 extra project rule files every auditor must apply; template=path
report template; no-project-lenses to skip the automatic CLAUDE.md +
.claude/rules/*.md ingestion.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimnpx claudepluginhub emasoft/emasoft-plugins --plugin code-auditor-agentTask distribution, agent coordination, progress monitoring - executes plans via subagents. Requires AI Maestro for inter-agent messaging.
Comprehensive validation, management, and standardization suite for Claude Code plugins and marketplaces. Includes 190+ validation rules, plugin lifecycle management, marketplace operations, health checks, security auditing, GitHub repo validation, plugin/marketplace repo scaffolding, and standardization tooling. Features severity hierarchy, --strict mode, language-aware token estimation, and universal plugin/marketplace templates.
GHE (GitHub-Elements) - Automated project management for Claude Code using GitHub Issues as persistent memory with orchestrated DEV/TEST/REVIEW workflow.
Portable utility tools for Claude Code plugin marketplaces. Includes release automation and markdown TOC generation.
Exports current session segment (since last compaction) with system-reminder stripping -- main conversation, subagent transcripts, sidechains, and debug logs in structured markdown
Multi-lens code review pipeline: deep review (Claude or Codex), automated fix loop, interactive walkthrough, manual promote, external-finding injection.
Automated code review for pull requests using multiple specialized agents with confidence-based scoring
Review pull requests with structured analysis and approve with confidence
Comprehensive PR review agents specializing in comments, tests, error handling, type design, code quality, and code simplification
Multi-agent code review for Claude Code — parallel review by Codex (GPT-5.5), Gemini 3.1 Pro, and five Claude specialist subagents (one run blind), then verified synthesis.
AI-powered code review in Claude Code, powered by CodeRabbit