docs: integrate veille hebdo Anthropic 17-23 fev 2026 (+177 lignes guide)

P1 - Prompt caching + env vars: - Strategy 6 cost optimization: cache_control API, pricing (write 1.25x, read 0.1x), break-even 2 hits - CLAUDE_CODE_DISABLE_1M_CONTEXT + CLAUDE_CODE_SIMPLE ajoutés à la table env vars principale P2 - Model deprecations + agents: - Warning claude-3-haiku-20240307 deprecated 19 fev / retirement 20 avril 2026 - Frontmatter fields: background: true (v2.1.49+) + isolation: worktree (v2.1.50+) - Section Background Agents: comportement non-blocking, gestion (ctrl+f, double ESC) - Section claude agents CLI: commande + output annoté P3 - Worktree isolation native + security cross-ref: - Native worktree features: --worktree flag, isolation declaratif, WorktreeCreate/WorktreeRemove hooks - §7.4 Security Hooks: cross-ref Claude Code Security scanner (research preview) Eval: docs/resource-evaluations/2026-02-22-veille-hebdo-anthropic-17-23-fev.md (score 3/5) Stats: 19K → 20K lignes (20 440 total) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-02-23 12:59:28 +01:00 · 2026-02-23 12:59:28 +01:00 · 6117145837
commit 6117145837
parent 2874eee382
8 changed files with 364 additions and 2 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -8,6 +8,26 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).

 ### Added

+- **guide/ultimate-guide.md** — 7 sections ajoutées suite à veille hebdo Anthropic 17-23 fev 2026 (éval: `docs/resource-evaluations/2026-02-22-veille-hebdo-anthropic-17-23-fev.md`, score 3/5)
+
+  **P1 — Prompt caching + env vars manquantes**
+  - §1.7 Cost Optimization — Strategy 6 : prompt caching API (`cache_control`, prix cache write/read, break-even, 4 breakpoints max)
+  - §Appendix Environment Variables — `CLAUDE_CODE_DISABLE_1M_CONTEXT` (v2.1.50+) et `CLAUDE_CODE_SIMPLE` (v2.1.50+) ajoutés à la table principale
+
+  **P2 — Model deprecations + agents**
+  - §1.7 Pricing Model — Warning box : `claude-3-haiku-20240307` deprecated 19 fev 2026, retirement 20 avril 2026, migration vers `claude-haiku-4-5-20251001`
+  - §4.2 Frontmatter Fields — `background: true` (non-blocking, v2.1.49+) et `isolation: "worktree"` (spawn isolé, v2.1.50+) ajoutés au tableau
+  - §4.4 Best Practices — Nouvelle section "Background Agents" : comportement, gestion (ctrl+f, double ESC/ctrl+C), comparatif mode default vs background
+  - §4.4 Best Practices — Nouvelle section "`claude agents` CLI" : commande + exemple d'output annoté (v2.1.50+)
+
+  **P3 — Worktree isolation native + cross-ref sécurité**
+  - §9.11 Git Worktrees — Nouvelle sous-section "Claude Code Native Worktree Features" : `--worktree`/`-w` flag, `isolation: "worktree"` déclaratif, hooks `WorktreeCreate`/`WorktreeRemove` avec exemple `settings.json` complet
+  - §7.4 Security Hooks — Callout cross-ref vers `security-hardening.md#claude-code-as-security-scanner-research-preview`
+
+- **docs/resource-evaluations/2026-02-22-veille-hebdo-anthropic-17-23-fev.md** — Fiche d'évaluation veille hebdo Anthropic 17-23 fev 2026 (score 3/5, intégration partielle)
+
+- **README.md + CLAUDE.md** — Mise à jour stat guide : `~19K lines` → `~20K lines` (20 440 lignes)
+
 - **guide/diagrams/09-cost-and-optimization.md + guide/ultimate-guide.md §2.5** — Budget modifier for model selection decision flow
  - Le diagramme "Model Selection Decision Flow" assumait implicitement un budget illimité (Max/API) — angle mort signalé par la communauté
  - Ajout d'un **Budget modifier table** : Max/API → Opus plan + Sonnet impl ; Pro/Teams Standard → Sonnet plan + Haiku impl (tâches mécaniques)
--- a/CLAUDE.md
+++ b/CLAUDE.md
@ -10,7 +10,7 @@ This repository is the **comprehensive documentation for Claude Code** (Anthropi

 ```
 guide/                    # Core documentation
-├── ultimate-guide.md     # Main guide (~19K lines, the reference)
+├── ultimate-guide.md     # Main guide (~20K lines, the reference)
 ├── cheatsheet.md         # 1-page printable summary
 ├── architecture.md       # How Claude Code works internally
 ├── methodologies.md      # TDD, SDD, BDD workflows
--- a/README.md
+++ b/README.md
@ -158,7 +158,7 @@ graph LR
 ```
 📦 claude-code-ultimate-guide/
 │
-├─ 📖 guide/              Core Documentation (~19K lines)
+├─ 📖 guide/              Core Documentation (~20K lines)
 │  ├─ ultimate-guide.md   Complete reference, 10 sections
 │  ├─ cheatsheet.md       1-page printable
 │  ├─ architecture.md     How Claude Code works internally
--- a/docs/resource-evaluations/2026-02-22-fp-dev-issue-tracker.md
+++ b/docs/resource-evaluations/2026-02-22-fp-dev-issue-tracker.md
@ -0,0 +1,108 @@
+# Resource Evaluation: fp.dev — Agent-Native Issue Tracking
+
+**Evaluated**: 2026-02-22
+**Evaluator**: Claude Sonnet 4.6 + technical-writer agent challenge
+**Target Guide**: Claude Code Ultimate Guide
+**Source**: https://fp.dev/ + LinkedIn post (community)
+
+---
+
+## Executive Summary
+
+**Resource**: fp.dev — "Agent-native issue tracking for ambitious Claude Code users"
+**URL**: https://fp.dev/
+
+**Initial Score**: 3/5
+**Challenged Score**: 2/5
+**Final Score**: **2/5 (Marginal — Watchlist)**
+
+**Decision**: **Ne pas intégrer activement. Ajouter dans Known Gaps + Watch List.**
+
+---
+
+## 📄 Résumé du contenu
+
+- **fp.dev** est un outil de tracking d'issues local-first conçu spécifiquement pour Claude Code
+- Issues stockées en `.md` dans `/.claude/` — git-committables (différentiateur réel vs Tasks API)
+- Workflow via skills custom (`/fp-plan`, `/fp-implement`, `/fp-review`) + CLI (`fp tree`, `fp issue list`, `fp issue show`, `fp review`)
+- Résout le "context rot" : permet de `/clear` la session sans perdre l'état projet
+- App desktop Mac (Apple Silicon uniquement) + UI code review avec diff viewer intégré
+- Zéro dépendance externe, zéro MCP server
+- Installation : `curl -fsSL https://setup.fp.dev/install.sh | sh -s && fp setup claude && fp init`
+
+---
+
+## 🎯 Score de pertinence (1-5)
+
+| Score | Signification |
+|-------|---------------|
+| 5 | Essentiel - Gap majeur dans le guide |
+| 4 | Très pertinent - Amélioration significative |
+| 3 | Pertinent - Complément utile |
+| 2 | Marginal - Info secondaire |
+| 1 | Hors scope - Non pertinent |
+
+**Score final : 2/5**
+
+**Justification** : Problème adressé réel mais déjà couvert partiellement par le guide (fresh-context-loop.sh, Tasks API native). Un seul différentiateur solide (issues .md git-committables). Adoption insuffisante, restriction Apple Silicon disqualifiante pour recommandation large.
+
+---
+
+## ⚖️ Comparatif
+
+| Aspect | fp.dev | Guide actuel |
+|--------|--------|-------------|
+| Context rot / session clear | ✅ Issues .md persistées | ✅ `fresh-context-loop.sh` (Ralph Wiggum pattern) |
+| Task management structuré | ✅ CLI + skills dédiés | ✅ Tasks API native |
+| Issues git-committables | ✅ (.md files dans repo) | ❌ Manquant |
+| Code review avec diff viewer | ✅ UI locale dédiée | ❌ Non couvert |
+| Multi-plateforme | ❌ Desktop = Apple Silicon only | ✅ N/A |
+| MCP intégration | ❌ Pas de MCP | ✅ Couvert dans le guide |
+| Équipes / sync | ❌ Local only | ✅ Tasks API multi-agent |
+
+---
+
+## 🔥 Challenge (technical-writer)
+
+Arguments pour descendre à 2/5 :
+
+1. **Redondance** : `/clear` + Tasks API couvre déjà le problème principal
+2. **Restriction plateforme** : Apple Silicon only pour le desktop = dealbreaker Linux/Windows
+3. **Signaux d'adoption manquants** : Pas de GitHub stars, pas de release cadence documentée → risque d'abandon
+4. **Mauvaise catégorisation** : Ne va pas aux côtés de Gas Town/multiclaude (multi-agent orchestration) — fp.dev est du task tracking
+5. **Un vrai différentiateur** : Issues `.md` git-committables → mais insuffisant seul pour 3/5
+
+**Score ajusté** : 2/5
+**Risques de non-intégration** : Faibles. Le guide couvre déjà le problème. La valeur est dans documenter le GAP (markdown-based issue tracking) pas nécessairement fp.dev.
+
+---
+
+## ✅ Fact-Check
+
+| Affirmation | Vérifiée | Source |
+|-------------|----------|--------|
+| Issues stockées en `.md` dans `/.claude/` | ✅ | fp.dev (WebFetch) + Perplexity |
+| CLI : `fp tree`, `fp issue list`, `fp issue show`, `fp review` | ✅ | WebFetch direct |
+| Installation : `fp setup claude` + `fp init` | ✅ | fp.dev + Perplexity |
+| Local-first, no external services | ✅ | fp.dev |
+| Mac desktop app = Apple Silicon requis | ✅ | fp.dev |
+| Skills `/fp-plan`, `/fp-implement`, `/fp-review` | ✅ | Description + WebFetch |
+
+Zéro hallucination. Toutes les claims vérifiées.
+
+---
+
+## 📍 Recommandations
+
+**Score < 3 → Watchlist, pas d'intégration directe.**
+
+Actions effectuées :
+1. ✅ Ajout dans `docs/resource-evaluations/watch-list.md`
+2. ✅ Ajout du gap "Agent-native issue tracking (markdown-based, git-committable)" dans `guide/third-party-tools.md` Known Gaps
+
+**Trigger de re-évaluation** : GitHub stars visibles + release cadence documentée + au moins un write-up praticien en production.
+
+---
+
+**Évaluation complétée** : 2026-02-22
+**Prochaine révision** : Sur trigger (adoption signals)
--- a/docs/resource-evaluations/2026-02-22-veille-hebdo-anthropic-17-23-fev.md
+++ b/docs/resource-evaluations/2026-02-22-veille-hebdo-anthropic-17-23-fev.md
@ -0,0 +1,55 @@
+---
+title: "Veille hebdomadaire Anthropic (17-23 fev 2026)"
+type: "weekly-watch"
+date: "2026-02-22"
+score: 3
+action: "partial-integration"
+sources:
+  - "Texte copie (digest agregateur, pas d'URL)"
+---
+
+# Veille hebdomadaire Anthropic (17-23 fev 2026)
+
+## Resume
+
+Digest agregateur couvrant Sonnet 4.6, Claude Code Security, CC v2.1.49-v2.1.50, Automatic Prompt Caching API, model deprecations, et activite Twitter.
+
+## Score: 3/5
+
+**Justification**: Digest de sources primaires (blog Anthropic, CHANGELOG GitHub, Twitter). 50% deja documente dans le guide. La valeur reelle = signal de 2-3 gaps : prompt caching API (zero couverture), model deprecations (zero couverture), env vars manquantes. Score 3 (baisse de 4 → 3 apres challenge) : c'est un trigger de recherche, pas une source citable — pas d'URL, claims API non verifiables directement.
+
+## Gaps identifies
+
+| Gap | Localisation | Priorite |
+|-----|-------------|----------|
+| Prompt caching API (`cache_control`) | guide/ultimate-guide.md (Cost Optimization) | **P1** |
+| `CLAUDE_CODE_DISABLE_1M_CONTEXT` env var | guide/ultimate-guide.md (env vars table) | **P1** |
+| `CLAUDE_CODE_SIMPLE` env var clarification | guide/ultimate-guide.md (env vars table) | **P1** |
+| Model deprecations (Haiku 3 → 20 avril 2026) | guide/ultimate-guide.md ou architecture.md | P2 |
+| `background: true` agent field | guide/ultimate-guide.md (agents section) | P2 |
+| `claude agents` CLI command | guide/ultimate-guide.md (agents section) | P2 |
+| Worktree hooks `WorktreeCreate`/`WorktreeRemove` | guide/ultimate-guide.md (worktree section) | P3 |
+
+## Fact-Check
+
+| Affirmation | Statut | Source |
+|-------------|--------|--------|
+| Sonnet 4.6 = default model | ✅ Verifie | ultimate-guide.md:1765 |
+| CC v2.1.49/v2.1.50 dates 19-20 fev | ✅ Verifie | claude-code-releases.md:27-51 |
+| Claude Code Security research preview | ✅ Verifie | security-hardening.md:780-832 |
+| Haiku 3 deprecated 19 fev 2026, retirement 20 avril | ✅ Verifie | Perplexity → platform.claude.com/docs/model-deprecations |
+| "Automatic" Prompt Caching (zero config) | ⚠️ Incorrecte | Anthropic utilise des breakpoints **explicites** (`cache_control`), pas automatiques. Perplexity confirme : "developer-controlled through explicit cache breakpoints, not automatic" |
+| Sonnet 3.7 / Haiku 3.5 retires | ⚠️ Non verifie | Pas retrouve dans sources Perplexity. A verifier sur docs.anthropic.com/release-notes |
+| Pricing prompt caching : 90% de reduction | ✅ Verifie | Cache reads = 0.1x base price (Perplexity + docs Anthropic) |
+
+## Actions P1 (effectuees)
+
+1. **Prompt caching** → Strategy 6 ajoutee dans Cost Optimization Strategies (ultimate-guide.md ~1912)
+2. **Env vars** → `CLAUDE_CODE_DISABLE_1M_CONTEXT` et `CLAUDE_CODE_SIMPLE` ajoutes a la table principale (ultimate-guide.md ~19862)
+
+## Challenge (technical-writer)
+
+- **Score ajuste** : 3/5 (baisse de 4 → 3)
+- **Raison** : Digest sans URL = source non citable. 50% deja couvert. La valeur = trigger de recherche primaire.
+- **Risque principal** : Claim "automatic prompt caching" est incorrect — Anthropic requiert `cache_control` explicite. Ne pas integrer cette formulation.
+- **Next steps** : Verifier Sonnet 3.7/Haiku 3.5 sur docs officiels avant integration P2.
--- a/docs/resource-evaluations/watch-list.md
+++ b/docs/resource-evaluations/watch-list.md
@ -16,6 +16,7 @@ Trigger reached → re-evaluation → Integrate (Graduated) / Drop (Dropped)

 | Resource | Type | Added | Why Watching | Re-eval Trigger |
 |----------|------|-------|--------------|-----------------|
+| [fp.dev](https://fp.dev/) | Tool | 2026-02-22 | Agent-native issue tracking pour Claude Code. Un vrai différentiateur (issues .md git-committables) mais adoption insuffisante, Apple Silicon only, redondant avec Tasks API. Score 2/5. | GitHub stars visibles + release cadence + write-up praticien en prod |
 | [ICM](https://github.com/rtk-ai/icm) | MCP | 2026-02-12 | Pre-v1 (1 star, 11 commits) | First release + >20 stars |
 | [System Prompts](https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools) | Tool | 2026-01-26 | Redundant with official sources. Re-evaluated 2026-02-13 (Opus 4.6 update): still 2/5. | Anthropic confirms CLI prompts not published |
 | [o16g — Outcome Engineering](https://o16g.com/) | Manifesto | 2026-02-13 | Emerging framework by Cory Ondrejka (CTO Onebrief, co-creator Second Life, ex-VP Google/Meta). 16 principles for shifting from "code writing" to "outcome engineering". Honeycomb endorsement. No Claude Code-specific content yet. Memetic potential (naming follows i18n/k8s pattern). | Term adopted in >3 independent AI engineering resources OR author publishes tool-specific implementation |
--- a/guide/third-party-tools.md
+++ b/guide/third-party-tools.md
@ -382,6 +382,7 @@ As of February 2026, the community tooling ecosystem has notable gaps:
 | **Visual hooks editor** | No GUI for managing hooks in `settings.json` — requires JSON editing |
 | **Unified admin panel** | No single dashboard combining config, sessions, cost, and MCP management |
 | **Session replay** | ✅ **FILLED**: Entire CLI (launched Feb 2026) provides rewindable checkpoints with full context replay |
+| **Agent-native issue tracking** | No established tool for markdown-based, git-committable issue tracking with Claude Code. [fp.dev](https://fp.dev/) is an early-stage solution (local-first, `/fp-plan` + `/fp-implement` skills, diff viewer) but lacks adoption signals and requires Apple Silicon for the desktop app. The Tasks API covers state persistence but issues aren't git-committable. |
 | **Per-MCP-server profiler** | No way to measure token cost attributable to each MCP server individually |
 | **Cross-platform config sync** | No tool syncs Claude Code config across machines (must manual copy `~/.claude/`) |

--- a/guide/ultimate-guide.md
+++ b/guide/ultimate-guide.md
@ -1771,6 +1771,8 @@ Claude Code uses **Claude Sonnet 4.6** by default (as of Feb 2026):

 **Reality check**: A typical 1-hour session costs **$0.10 - $0.50** depending on usage patterns.

+> **Model deprecations (Feb 2026)**: `claude-3-haiku-20240307` (Claude 3 Haiku) was deprecated on **February 19, 2026** with **retirement scheduled for April 20, 2026**. If your CLAUDE.md, agent definitions, or scripts hardcode this model ID, migrate to `claude-haiku-4-5-20251001` (Haiku 4.5) before April 2026. Source: [platform.claude.com/docs/model-deprecations](https://platform.claude.com/docs/model-deprecations)
+
 #### 200K vs 1M Context: Performance, Cost & Use Cases

 The 1M context window (beta, API + usage tier 4 required) is a significant capability jump — but community feedback consistently frames it as a **niche premium tool**, not a default.
@ -1911,6 +1913,45 @@ claude --model opus "Design the entire authentication system"
 # Shared context, single response
 ```

+**Strategy 6: Use prompt caching for repeated context (API)**
+
+If you call the Anthropic API directly (e.g., for custom agents or pipelines), prompt caching cuts costs by up to 90% on repeated prefixes.
+
+```python
+# Mark stable sections with cache_control
+response = client.messages.create(
+    model="claude-sonnet-4-6-20250514",
+    max_tokens=1024,
+    system=[
+        {
+            "type": "text",
+            "text": "<your large system prompt / codebase context>",
+            "cache_control": {"type": "ephemeral"}  # Cache this prefix
+        }
+    ],
+    messages=[{"role": "user", "content": "Fix the bug in auth.ts"}]
+)
+```
+
+**Prompt caching economics**:
+
+| Operation | Cost multiplier | TTL |
+|-----------|-----------------|-----|
+| Cache write | 1.25x base price | 5 minutes (default) |
+| Cache write (extended) | 2x base price | 1 hour |
+| Cache read (hit) | 0.1x base price | — |
+| Latency reduction | Up to 85% for long prompts | — |
+
+**Break-even**: 2 cache hits with 5-minute TTL. After that, pure savings.
+
+**Rules**:
+- Max **4 cache breakpoints** per request
+- Cache key = exact prefix match (single character change = cache miss)
+- Place breakpoints after large stable sections: system prompt, tool definitions, codebase context
+- For Claude Code itself: caching is handled automatically by the CLI — this applies to API-based workflows you build on top of Claude
+
+> Docs: [prompt caching](https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching)
+
 #### Tracking Costs

 **Real-time tracking**:
@ -5500,6 +5541,8 @@ disallowedTools:
 | `skills` | ❌ | Skills to inherit |
 | `disallowedTools` | ❌ | Tools to block |
 | `memory` | ❌ | Pre-populated memory context (v2.1.32+) |
+| `background` | ❌ | Always run as background task — non-blocking (v2.1.49+) |
+| `isolation` | ❌ | `"worktree"` — spawn agent in isolated git worktree (v2.1.50+) |

 **New in v2.1.32**: The `memory` field enables pre-populated agent context:

@ -5663,6 +5706,64 @@ Before deploying a custom agent, validate against these criteria:

 > **Automated audit**: Run `/audit-agents-skills` for a comprehensive quality audit across all agents, skills, and commands. Scores each file on 16 criteria with weighted grading (32 points for agents/skills, 20 for commands). See `examples/skills/audit-agents-skills/` for the full scoring methodology.

+### Background Agents (v2.1.49+)
+
+The `background: true` field makes an agent always run non-blocking — the parent session continues immediately instead of waiting for the agent to finish.
+
+```yaml
+---
+name: test-runner
+description: Run the full test suite after every significant change
+model: sonnet
+tools: Bash
+background: true   # Non-blocking — parent continues without waiting
+---
+
+Run `npm test` and report results. Flag failures with file + line number.
+```
+
+**Vs. standard agents:**
+
+| Mode | Behavior | Use when |
+|------|----------|----------|
+| Default | Parent waits for agent output | Need result before continuing |
+| `background: true` | Agent runs in parallel, parent continues | Fire-and-forget (tests, linting, notifications) |
+
+**Managing background agents:**
+
+```bash
+# List running agents
+ctrl+f    # Opens agent manager overlay
+
+# Kill all background agents (double-press within 3s)
+ctrl+c ctrl+c
+# or
+ESC ESC
+```
+
+### `claude agents` CLI (v2.1.50+)
+
+List all configured agents across global and project directories:
+
+```bash
+claude agents
+```
+
+**Example output:**
+
+```
+Global agents (~/.claude/agents/):
+  ├── code-reviewer   [sonnet] — Code quality reviews, security audits
+  ├── test-runner     [sonnet] [background] — Run full test suite
+  └── security-audit  [opus]   — OWASP Top 10, auth, encryption
+
+Project agents (.claude/agents/):
+  ├── backend-arch    [opus]   — API design, database schemas
+  └── docs-writer     [haiku]  — Generate docs from code
+```
+
+Useful for debugging agent discovery issues or auditing which agents are available in a project.
+
 ## 4.5 Agent Examples

 ### Example 1: Code Reviewer Agent
@ -8130,6 +8231,8 @@ Security hooks are critical for protecting your system.

 > **Advanced patterns**: For comprehensive security including Unicode injection detection, MCP config integrity verification, and CVE-specific mitigations, see [Security Hardening Guide](./security-hardening.md).

+> **Claude Code Security (research preview)**: Anthropic offers a dedicated codebase vulnerability scanner that traces data flows across files, challenges findings internally before surfacing them (adversarial validation), and generates patch suggestions. Separate from the Security Auditor Agent above — waitlist access only. See [Security Hardening Guide → Claude Code as Security Scanner](./security-hardening.md#claude-code-as-security-scanner-research-preview).
+
 ### Recommended Security Rules

 ```bash
@ -13824,6 +13927,78 @@ git worktree remove --force .worktrees/old-feature
  - [`examples/commands/git-worktree-remove.md`](../examples/commands/git-worktree-remove.md) — Remove
  - [`examples/commands/git-worktree-clean.md`](../examples/commands/git-worktree-clean.md) — Clean

+### Claude Code Native Worktree Features (v2.1.49–v2.1.50)
+
+Claude Code has built-in worktree integration beyond the manual `git worktree` workflow above.
+
+#### Start Claude in an isolated worktree
+
+```bash
+# --worktree / -w flag: creates a temporary worktree based on HEAD
+claude --worktree
+claude -w
+```
+
+The worktree is created automatically, Claude runs inside it, and it is cleaned up on exit (if no changes were made).
+
+#### Declarative isolation in agent definitions
+
+Set `isolation: "worktree"` in an agent's frontmatter to automatically spawn it in a fresh worktree every time (v2.1.50+):
+
+```yaml
+---
+name: refactoring-agent
+description: Large-scale refactors that must not pollute the main working tree
+model: opus
+isolation: "worktree"   # Each invocation gets its own isolated checkout
+---
+
+Perform the requested refactoring. Commit your changes inside the worktree.
+```
+
+This replaces the earlier pattern of manually passing `isolation: "worktree"` to each Task tool call.
+
+#### Custom VCS setup with hook events (v2.1.50+)
+
+Two new hook events fire around agent worktree lifecycle:
+
+| Event | Fires | Use case |
+|-------|-------|----------|
+| `WorktreeCreate` | When an agent worktree is created | Set up DB branch, copy .env, install deps |
+| `WorktreeRemove` | When an agent worktree is torn down | Clean up DB branch, delete temp credentials |
+
+```json
+// .claude/settings.json
+{
+  "hooks": {
+    "WorktreeCreate": [
+      {
+        "matcher": "",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "scripts/worktree-setup.sh $CLAUDE_WORKTREE_PATH"
+          }
+        ]
+      }
+    ],
+    "WorktreeRemove": [
+      {
+        "matcher": "",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "scripts/worktree-teardown.sh $CLAUDE_WORKTREE_PATH"
+          }
+        ]
+      }
+    ]
+  }
+}
+```
+
+Typical `worktree-setup.sh`: create a Neon/PlanetScale DB branch, copy `.env.local`, run `npm install`.
+
 ### Database Branch Isolation with Worktrees

 **Modern pattern (2024+):** Combine git worktrees with database branches for true feature isolation.
@ -19860,6 +20035,8 @@ Set these in your shell profile (`~/.zshrc`, `~/.bashrc`, or Windows System Prop
 | `ANTHROPIC_SMALL_FAST_MODEL` | Fast model for simple tasks | `claude-haiku-4-20250514` |
 | `BASH_DEFAULT_TIMEOUT_MS` | Bash command timeout | `60000` |
 | `ANTHROPIC_AUTH_TOKEN` | Alternative auth token | Your auth token |
+| `CLAUDE_CODE_DISABLE_1M_CONTEXT` | Disable 1M context window support (v2.1.50+) | `true` |
+| `CLAUDE_CODE_SIMPLE` | Fully minimal mode: disables skills, agents, MCP, hooks, CLAUDE.md loading (v2.1.50+) | `true` |

 ### Finding Your Paths