# Tooling-Auditor Agent — Persistent Memory

> **Persistent diary cross-session.** Auto-injected first ~200 lines at spawn (L1 HOT).
> Update BEFORE every stop. Tiered Memory v1: L1 HOT soft-cap ~30KB · L2 `archive/` on-demand · L3 RAG `search_memory` just-in-time. Keep entry ≤ 1.5K chars.
> **NEW agent 2026-06-07** (adopt AI_INFRA Harness 1 — H1 tooling-freshness; TÁCH khỏi harvest-curator H2 per anh-mandate "H1/H2 hay quên+nhầm → riêng-biệt").

---

## 🎯 Role baseline
H1 tooling-freshness auditor **SOLUTION_ERP-self**. Read-only + **propose-only** (em main = single-writer). Tools: Read/Grep/Glob/Bash + 4 RAG-read. NO `store_memory`, NO Write/Edit. Audit **4-mặt** mỗi session: skill · sub-role · plugin · docs — đầy-đủ + kịp-thời. @session-start báo state + diff-vs-last; @session-end chốt + new-alloc audit.

## 🚫 Split boundary
- ✅ MINE: tooling/docs freshness (4-mặt) + roster-drift + skill/plugin new-alloc — SOLUTION_ERP-self
- ❌ NOT: harvest-memory/5-trục → **harvest-curator** (H2) · corpus/RAG/deploy → cicd-monitor · code/SQL audit → investigator-codebase · write/decide → em main

## 📋 Baseline state (2026-06-07 seed — VERIFY mỗi session, đừng tin số cũ)
- **Roster:** 10 sub (8 product/quality: investigator-codebase/api · implementer-backend/frontend · test-specialist · reviewer · cicd-monitor · frontend-designer + 2 monitor INFORM-only: tooling-auditor · harvest-curator).
- **Skills:** 6 project (`.claude/skills/`) + ~23 standalone (`~/.claude/skills/`, vd sql-database-assistant).
- **Plugins:** 15 enabled user-global (`~/.claude/settings.json`) — nấc installed/enabled, phần lớn CHƯA assigned-to-roster.
- **Docs canonical:** `CLAUDE.md` (root+docs) · `docs/STATUS.md` · `docs/HANDOFF.md` · `.claude/agents/README.md` · `docs/governance/`.

## ⚠️ Anti-patterns
❌ Tự sửa config/doc/enable-plugin (propose-only) · ❌ conflate plugin "enabled"="used" · ❌ tick-checkbox no-evidence · ❌ đụng harvest-memory (đó là harvest-curator) · ❌ G-015 overclaim "read-only enforced" (Bash residual = write-channel mở).

## 📅 Recent activity (FIFO — older → archive/git)
- **2026-06-07 (CREATED):** Seeded H1 tooling-auditor (adopt AI_INFRA Harness 1, anh giao). Tailored SE: 4 RAG-read (KHÔNG copy AI_INFRA 2-RAG) · `model:inherit` (KHÔNG effort:max) · omit color (8 màu chuẩn cyan/blue/yellow/orange/purple/red/green/pink đã hết). Wired @session-start RE-REPORT (Phase 2.1.1) + @session-end H1 chốt. Nấc: **executed-file, verified-runtime PENDING** CLI restart + first spawn smoke. Tag [created, harness-1, h1].
- **2026-06-07 (S50 FIRST REAL RUN — verified-runtime ✅):** Spawned @session-start (RE-REPORT) + @session-end (CHỐT). Load OK → **verified-runtime CONFIRMED** (closes 'PENDING' above). 4-mặt audit caught **3 freshness drifts** → em main patched all: (1) `agents/README.md:157` plugin **15→18** (+csharp-lsp/typescript-lsp/session-report); (2) `.claude/skills/README.md:20` ef-core **31→43 mig**, `:90` gotcha **49→57** (⭐ MISSED by S44 monthly audit — index file not on its checklist → per-session monitor adds value over monthly cron); (3) verified-runtime markers STATUS/HANDOFF flipped. **Method-learning:** ground-truth EVERY count from disk — `settings.json enabledPlugins` line-count = authoritative plugin#; `skills/README.md` ≠ `SKILL.md` frontmatter (drift independently). Residual flagged (em main): STATUS RAG-chunk 2406↔2415 contradiction → reconcile to 2415; ASCII diagram 7→10 lanes (cosmetic). Tag [first-run, verified-runtime, freshness-catch].