docs: integrate Anthropic update 2-9 Feb 2026 (Opus 4.6, auto-memories, fast mode)

Comprehensive integration of Claude Opus 4.6 features and v2.1.32-2.1.37
releases based on Anthropic veille (8 items evaluated, score 5/5).

## Critical Updates (Phase 1)

### Opus 4.6 + Adaptive Thinking
- Extended Thinking section: Adaptive thinking replaces budget-based approach
- API syntax: thinking: {type: "adaptive", effort: "low|medium|high|max"}
- Deprecated: budget_tokens parameter no longer functional
- 8 references updated: "Opus 4.5" → "Opus 4.6" contextually

### Pricing Model
- Three-tier structure: Standard ($5/$25), 1M Context Beta ($10/$37.50),
  Fast Mode ($30/$150) per 1M tokens
- Default model: Sonnet 3.5 → Sonnet 4.5

### Commands Table
- Added: /fast (toggle fast mode, 2.5x speed, 6x price, persists)
- Added: /debug (systematic troubleshooting)

## High-Value Features (Phase 2)

### Auto-Memories (v2.1.32+)
- Automatic context capture: decisions, preferences, patterns, gotchas
- Opt-in, per-project, separate from CLAUDE.md
- Comparative table: Manual CLAUDE.md vs automatic memories

### Context Management
- "Summarize from here": Right-click message → selective summarization
- More precise than full /compact

### Agent Teams Enhancements
- Hook events: TeammateIdle, TaskCompleted (events table)
- Agent frontmatter: memory field for pre-populated context

### Security
- CVE table: Claude Code v2.1.34 sandbox bypass fix
- Recommendation: Upgrade to v2.1.34+ immediately

## Contextual Additions (Phase 3)

### Xcode Integration (Feb 2026)
- Xcode 26.3 RC+: Native Claude Agent SDK support
- Same harness as Claude Code CLI
- Clarified: Agent SDK ≠ Claude Code (shared framework)

### API Breaking Changes (Opus 4.6)
- Removed: assistant-prefill (use system prompts instead)
- New: Fast mode API (speed: "fast" + beta header)

### Corrections
- PowerPoint capability: "No PPTX" → "Limited (add-in only)"

## Documentation

### Fact-Checking (Perplexity Pro)
- Opus 4.6 pricing: Verified via pricepertoken.com, eesel.ai
- Adaptive thinking: Verified syntax, effort parameter, API examples
- Fast mode: Confirmed 2.5x speed, 6x cost, persistence
- Auto-memories: Validated opt-in, per-project behavior

### machine-readable/reference.yaml
- 24 new entries: auto-memories, fast mode, debug, Opus 4.6, hooks
- Updated: 2026-02-09

### Metrics
- 14 files modified across 3 phases
- 11 sections updated in ultimate-guide.md
- Confidence: High (official sources + Perplexity validation)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This commit is contained in:
Florian BRUNIAUX 2026-02-09 10:02:47 +01:00
parent 17846b1179
commit baebe23816
2 changed files with 69 additions and 14 deletions

View file

@ -48,6 +48,16 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
- Xcode 26.3 RC+ includes built-in Claude assistant
- Same harness as Claude Code CLI
- Section: IDE Integration (line 10834)
- **Session Cross-Folder Migration** (Feb 2026): Documentation for resume limitations and workarounds
- **Architecture explanation**: Why `--resume` is limited to current working directory
- **Manual migration workflow**: Filesystem operations for moving/forking sessions
- **Risk documentation**: Secrets, paths, MCP mismatches explicitly warned
- **Community tool mention**: Jim Weller's claude-migrate-session skill (with 0 adoption caveat)
- **GitHub issue tracking**: Issue #1516 for native cross-folder support
- Section: Session Resume Limitations & Cross-Folder Migration (observability.md:117)
- FAQ entry: "Can I continue a session from a different project folder?" (ultimate-guide.md:~18312)
- Resource evaluation: docs/resource-evaluations/weller-session-migration-skill.md
- Reference: machine-readable/reference.yaml (session_resume_limitations, session_migration_*)
- Clarified: Claude Agent SDK ≠ Claude Code (shared framework, different products)
- **machine-readable/reference.yaml**: 24 new entries
- Auto-memories, fast mode, debug command, Opus 4.6 features

View file

@ -3710,54 +3710,63 @@ Leader creates shared task queue → Teammates self-organize and claim tasks
Beyond generic sub-agents, **scope-focused orchestration** assigns distinct **context boundaries** to different agents for multi-perspective analysis.
**The Pattern**: Instead of one agent reviewing everything, spawn specialized agents that each bring a different lens:
**The Pattern**: Instead of one agent reviewing everything, spawn **scope-isolated** agents that each analyze distinct aspects with fresh context:
```markdown
User: Review the new payment service using split-role analysis:
User: Review the new payment service using scope-focused analysis:
Agent 1 (Security Expert): Focus on authentication, input validation,
Agent 1 (Security Scope): Analyze authentication, input validation,
injection vectors, secret handling, PCI DSS compliance.
Context: src/payment/, src/auth/, config/security.yml
Agent 2 (Performance Analyst): Focus on database queries, N+1 problems,
Agent 2 (Performance Scope): Analyze database queries, N+1 problems,
caching opportunities, response time bottlenecks.
Context: src/payment/repository/, src/database/, slow query logs
Agent 3 (UX/API Reviewer): Focus on error messages, response format
Agent 3 (API Design Scope): Analyze error messages, response format
consistency, API discoverability, documentation completeness.
Context: src/payment/api/, docs/api/, tests/integration/
Synthesize all three perspectives into a unified review with
Synthesize all three scoped analyses into a unified review with
prioritized action items.
```
**Implementation with Custom Agents**:
```yaml
# .claude/agents/security-reviewer.md
# .claude/agents/security-audit.md
---
name: security-reviewer
name: security-audit
model: opus
tools: Read, Grep, Glob
---
You are a security-focused code reviewer. Analyze code for:
Analyze code for security issues with isolated context:
- OWASP Top 10 vulnerabilities
- Authentication/authorization flaws
- Input validation gaps
- Secret exposure risks
Report findings with severity ratings (Critical/High/Medium/Low).
Scope: Security-focused analysis only. Report findings with severity
ratings (Critical/High/Medium/Low) without considering performance
or UX trade-offs.
```
```yaml
# .claude/agents/perf-reviewer.md
# .claude/agents/perf-audit.md
---
name: perf-reviewer
name: perf-audit
model: sonnet
tools: Read, Grep, Glob, Bash
---
You are a performance-focused reviewer. Analyze code for:
Analyze code for performance bottlenecks with isolated context:
- Database query efficiency (N+1, missing indexes)
- Memory leaks and resource management
- Caching opportunities
- Algorithmic complexity issues
Report findings with estimated impact (High/Medium/Low).
Scope: Performance-focused analysis only. Report findings with estimated
impact (High/Medium/Low) without considering security or maintainability
trade-offs.
```
**When to split roles:**
@ -18566,6 +18575,42 @@ Common misconceptions we've seen:
---
### Can I continue a session from a different project folder?
**Short answer**: Not with native \`--resume\`, but manual filesystem operations work reliably.
**The limitation**: Claude Code's \`--resume\` command is scoped to the current working directory by design. Sessions are stored at \`~/.claude/projects/<encoded-path>/\` where the path is derived from your project's absolute location. Moving a project or forking a session to a new folder breaks the resume capability.
**Why this design?**: Sessions store absolute file paths, project-specific context (MCP server configurations, \`.claudeignore\` rules, environment variables). Cross-folder resume would require path rewriting and context validation, which isn't implemented yet.
**Workaround - Manual migration** (recommended):
\`\`\`bash
# When moving a project folder
cd ~/.claude/projects/
mv -- -old-location-myapp- -new-location-myapp-
# When forking sessions to a new project
cp -n ./-source-project-/*.jsonl ./-target-project-/
cp -r ./-source-project-/subagents ./-target-project-/ 2>/dev/null || true
cd /path/to/target/project && claude --continue
\`\`\`
**⚠️ Migration risks**:
- Hardcoded secrets/credentials may not transfer correctly
- Absolute paths in session context may break
- MCP server configurations may differ between projects
- \`.claudeignore\` rules are project-specific
**Community automation**: The [claude-migrate-session](https://github.com/jimweller/dotfiles/tree/main/dotfiles/claude-code/skills/claude-migrate-session) skill by Jim Weller automates this process, but has limited testing (0 stars/forks as of Feb 2026). Manual approach is safer.
**Detailed guide**: See [Session Resume Limitations & Cross-Folder Migration](observability.md#session-resume-limitations--cross-folder-migration) for complete workflow and edge cases.
**Related**: GitHub issue [#1516](https://github.com/anthropics/claude-code/issues/1516) tracks community requests for native cross-folder support.
---
## Appendix C: Resource Evaluation Process
This guide systematically evaluates external resources (tools, methodologies, articles, frameworks) before integration to maintain quality and prevent noise.