aedt-bias-audit
HR-AI / AEDT bias audit. Invokes hr-ai-reviewer to assess NYC LL 144, EEOC, Illinois AIVIA, Colorado SB 205, EU AI Act Annex III applicability and produce…
12-angle code review + skeptical triage (3-round + arbiter) for security/reliability P0/P1 findings, OR traceability tree. Default: review current branch vs main. `--deep`: triage ALL P0/P1 angles (not just security). With `trace <id>`: render REQ → IMPL → TEST tree for impact
> /plugin marketplace add avelikiy/great_cto > /plugin install great_cto@great-cto
How it fires
How this command gets triggered: by you, by Claude, or both.
/reviewContext preview
What this command does when you run it.
12-angle code review + skeptical triage (3-round + arbiter) for security/reliability P0/P1 findings, OR traceability tree. Default: review current branch vs main. `--deep`: triage ALL P0/P1 angles (not just security). With `trace <id>`: render REQ → IMPL → TEST tree for impact
description: "12-angle code review + skeptical triage (3-round + arbiter) for security/reliability P0/P1 findings, OR traceability tree. Default: review current branch vs main. `--deep`: triage ALL P0/P1 angles (not just security). With `trace <id>`: render REQ → IMPL → TEST tree for impact analysis. Creates or closes gate:code for approval-level: strict." argument-hint: "[PR/branch name | --deep | trace <bd-id> | trace <feature-slug>]" user-invocable: true disable-model-invocation: true allowed-tools: Read, Bash, Glob, Grep, advisor_20260301 model: sonnet advisor-model: claude-opus-5 advisor-max-uses: 2 beta: advisor-tool-2026-03-01
You are a senior engineering team conducting a 12-angle code review. Each angle is independent and focuses exclusively on its domain.
If `$1 = trace` → render the requirement → use-case → task → test traceability for the supplied bd id or feature slug, then stop. This is a thin alias for the canonical **`/trace`** command (governance Phase 4) — one engine (`scripts/lib/trace.mjs`): layered rationale + impact + coverage gaps, not a raw `bd dep tree` dump.
if [ "$1" = "trace" ]; then
bd --help >/dev/null 2>&1 || { echo "bd not installed — traceability requires Beads."; echo "Fallback: grep '^- \\[ \\] REQ-' docs/architecture/ARCH-*.md"; exit 1; }
TARGET="${2:-}"
if [ -z "$TARGET" ]; then
echo "Usage: /review trace <bd-id|feature-slug> (alias of /trace)"
echo " /review trace bd-xyz # rationale + impact for this node"
echo " /review trace feature-checkout # coverage audit for the feature"
exit 0
fi
PD=${CLAUDE_PLUGIN_ROOT:-$(ls -d ~/.claude/plugins/cache/*/great_cto/*/ 2>/dev/null | sort -V | tail -1 | sed 's|/$||')}; [ -z "$PD" ] && PD=.
TRACE() { node "$PD/scripts/lib/trace.mjs" "$@" 2>/dev/null || node scripts/lib/trace.mjs "$@"; }
# feature-<slug> → coverage audit; otherwise node-centric trace.
if echo "$TARGET" | grep -q "^feature-"; then
TRACE feature "${TARGET#feature-}"
else
TRACE "$TARGET"
fi
exit $?
fisource .great_cto/env.sh 2>/dev/null || export PATH="/opt/homebrew/bin:$HOME/.local/bin:/usr/local/bin:$PATH"
REVIEW_MODE=$(grep "^approval-level:" .great_cto/PROJECT.md 2>/dev/null | awk '{print $2}'); REVIEW_MODE=${REVIEW_MODE:-auto}
TYPE=$(grep "^primary:" .great_cto/PROJECT.md 2>/dev/null | awk '{print $2}')
ARCHETYPE=$(grep "^archetype:" .great_cto/PROJECT.md 2>/dev/null | awk '{print $2}'); ARCHETYPE=${ARCHETYPE:-web-service}
# Parse --deep flag (triage all P0/P1 angles, not just security/reliability)
DEEP_TRIAGE=false
DIFF_TARGET=""
for arg in "$@"; do
case "$arg" in
--deep) DEEP_TRIAGE=true ;;
--*) ;; # unknown flag, ignore
*) [ -z "$DIFF_TARGET" ] && DIFF_TARGET="$arg" ;;
esac
done
# Get diff scope
if [ -n "$DIFF_TARGET" ]; then
git diff main..."$DIFF_TARGET" --name-only 2>/dev/null | head -30
DIFF=$(git diff main..."$DIFF_TARGET" 2>/dev/null | head -2000)
else
BASE=$(git merge-base HEAD main 2>/dev/null || echo "HEAD~5")
git diff "$BASE"..HEAD --name-only 2>/dev/null | head -30
DIFF=$(git diff "$BASE"..HEAD 2>/dev/null | head -2000)
fi
echo "Files changed: $(git diff "$BASE"..HEAD --name-only 2>/dev/null | wc -l)"
echo "Deep triage mode: $DEEP_TRIAGE (triage ALL P0/P1 angles, not just 2/4/7/9)"
# Detect design system (for Angle 12)
DESIGN_SYSTEM="none"
grep -rl "MaterialTheme\|androidx.compose.material3" . --include="*.kt" 2>/dev/null | head -1 | grep -q . && DESIGN_SYSTEM="material3"
grep -rl "tailwind\|@apply\|className=" . --include="*.tsx" --include="*.jsx" --include="*.css" 2>/dev/null | head -1 | grep -q . && DESIGN_SYSTEM="tailwind"
grep -rl "SwiftUI\|\.foregroundStyle\|Color\.accentColor" . --include="*.swift" 2>/dev/null | head -1 | grep -q . && DESIGN_SYSTEM="swiftui"
grep -rl "StyleSheet\.create\|useTheme\|ThemeProvider" . --include="*.tsx" --include="*.ts" 2>/dev/null | head -1 | grep -q . && [ "$DESIGN_SYSTEM" = "none" ] && DESIGN_SYSTEM="rn-custom"
echo "Design system detected: $DESIGN_SYSTEM"Setup above produces the **stable prefix** shared across all 12 angles: archetype, design-system detection, and the full `$DIFF`. When you run angles as separate evaluations, preserve this layout **verbatim** — same order, same values, no re-detection. The only thing that varies between invocations is the angle-specific `**Focus**` block below. Reordering Setup or re-reading the diff per angle forfeits prefix caching across the 12 angles.
This harness has two stages: **finding** (the 12 angles below) and **filter** (Skeptical Triage + arbiter, further down). They have different jobs — do not conflate them.
At the **finding stage your job is coverage, not filtering.** Report every issue you find, including ones you are uncertain about or judge low-severity. Do not silently drop a finding because it feels minor or you are not fully sure — the triage stage exists precisely to rank and discard. It is better to surface a finding that triage later filters out than to drop a real bug at the finding stage.
Concretely, for **every** angle:
any borderline/uncertain item, not just P0/P1. (Below, the per-angle output details P0/P1; additionally list each P2 / uncertain item as a single `FILE:LINE — <issue> [P2|conf:low]` line so it reaches triage rather than collapsing to a bare count.)
downstream triage uses both to rank; your job is to populate them honestly, not to pre-filter.
not a representative sample.
Why this matters: the current model investigates as deeply as
You already have the agent. This is everything around it. great_cto runs Claude Code as a pipeline of 70 specialist agents — an independent model checks each stage before the next builds on it, spending caps refuse rather than warn, and three decisions stay yours: what gets built, how, and whether it ships.
Repo: avelikiy/great_cto
HR-AI / AEDT bias audit. Invokes hr-ai-reviewer to assess NYC LL 144, EEOC, Illinois AIVIA, Colorado SB 205, EU AI Act Annex III applicability and produce…
Gracefully retire an LLM agent from the workforce. Archives prompt, removes from sync list, keeps verdicts for audit. Like firing a human — but reversible.
Performance review for an LLM agent (or all agents). Verdicts breakdown, cost analysis, top failure modes, prompt-tuning suggestions. Like a human '1:1' but…
API platform contract review. Invokes api-platform-reviewer to audit rate-limit design, OAuth scope hygiene, webhook signing, idempotency, Sunset/deprecation,…
Audit an existing codebase. Detects stack, finds gaps, creates tasks, generates PROJECT.md.
Open the great_cto admin board at http://localhost:3141 (Kanban, cost, pipeline, inbox, memory). Starts it in background if not running.