Competitive Audit

Last updated: 2026-03-25 (Item #10) Next review: Weekly (via analyze-community.md competitive watchlist in weekly community scan)

Purpose

Honest assessment of where we stand in the Claude Code ecosystem. What we do well, what others do better, and what we should incorporate or share back. Open source spirit — we're allies, not enemies.

Ecosystem Overview

Project	Stars	Focus	Relationship
everything-claude-code	50K+	Full agent framework (28 agents, 80+ skills)	Different niche — breadth vs our depth
claude-sdlc	—	Plugin marketplace (13 commands, 3 agents)	Closest competitor — similar philosophy
claude-code-sdlc	—	12-agent documentation-first pipeline	Documentation overlap
awesome-claude-code	31.9K	Curated resource directory	Community pulse
aistupidlevel.info	—	Model benchmarking (CUSUM, CI, drift)	Methodology ally — inspired our stats.sh

What We Do That Nobody Else Does

1. Statistical E2E evaluation — 95% confidence intervals with t-distribution, 5-trial Tier 2 testing. No other Claude Code tool measures itself this rigorously.
2. SDP scoring — Separates "the model had a bad day" (L1) from "our SDLC broke" (L2) using external benchmark cross-referencing. Formula: SDP = Raw x (baseline_external / current_external).
3. CI Shepherd — Local shepherd loop watches CI, reads failure logs, fixes locally, and pushes — all within one session with full context. Max 2 fix attempts before escalating to user.
4. Prove-It A/B pipeline — Before keeping any custom feature, we prove it outperforms native alternatives with statistical evidence. If we can't prove it, we delete it.
5. CUSUM drift detection — Per-criterion tracking (not just overall). Catches "plan_mode is degrading while tdd_green is stable" — isolates which SDLC area is slipping.
6. Dogfooding enforcement — The wizard tests itself. Every PR to this repo runs the same SDLC evaluation we recommend for users. Comprehensive automated test suite validated by CI on every PR.
7. Auto-update research — Weekly Claude Code release checks + community scan, monthly deep research. Automated trend detection.

What Others Do Better

Feature	Who	What They Do	Our Status
Language coverage	everything-claude-code	7+ languages, 10 ecosystems	Python/JS-centric (by design)
Skill library	everything-claude-code	80+ domain-specific skills	4 skills (by design — minimal)
Cross-platform hooks	everything-claude-code	Node.js (Windows/macOS/Linux)	Bash only (macOS/Linux)
Token/cost tracking	everything-claude-code	Per-session cost monitoring	Blocked — claude-code-action doesn't expose usage data
Webhook automation	claude-sdlc	GitHub issues auto-trigger SDLC	Manual PR creation
npm distribution	everything-claude-code, claude-sdlc	npm install or GitHub App	npx agentic-sdlc-wizard init (CLI shipped)
Pattern learning	everything-claude-code	Auto-extract patterns from sessions	Manual identification

What We Incorporated From This Audit

aistupidlevel.info Integration

• Before: Cited methodology but never fetched their data
• After: Added as Source 3 in external benchmark cascade (DailyBench -> LiveBench -> aistupidlevel -> baseline)
• API: https://aistupidlevel.info/api/dashboard/scores — JSON with model scores, confidence intervals, trend data
• Value: Third independent data source for SDP scoring. Their confidence intervals validate our CI calculations.

Competitive Watchlist

• Before: Weekly community scan checked general discussions
• After: analyze-community.md now includes named repos to check for new releases/patterns
• Repos tracked: everything-claude-code, claude-sdlc, claude-code-sdlc, awesome-claude-code, aistupidlevel

What We Could Contribute Back

Contribution	Target	Value
Per-criterion CUSUM	aistupidlevel.info	Granular drift detection (they track overall only)
SDP two-layer scoring	aistupidlevel.info	Separating model quality from task compliance
Self-healing CI pattern	Community	Auto-fix + re-trigger loop
Prove-It A/B methodology	Community	Statistical comparison framework for AI tools
stats.sh library	Community	Tested t-distribution 95% CI calculation in bash

Status: Documented for future. Will open issues/PRs when timing is right.

Tracked Gaps

Token Usage Tracking (Blocked)

• Problem: If an update causes 3x more tokens, that's a real regression even if SDLC score holds
• Blocker: claude-code-action@v1 doesn't expose usage data in execution output
• When: Weekly update workflow will detect when this becomes available
• Plan: Add tokens/run to score-history.jsonl, token regression detection to compare_ci

Distribution (Shipped)

• CLI install shipped: npx agentic-sdlc-wizard init
• Remaining: docs polish, npm registry publishing

Philosophical Position

We're not competing with everything-claude-code on breadth. We're a minimal, focused SDLC baseline that:

• Ships in a single wizard file + hooks
• Proves itself with statistical evidence
• Deletes custom code when native features catch up
• Gets better automatically through self-evolving research

The goal isn't to be the biggest — it's to be the most honest about whether it works.