CC 2.1.84 Adoption Brainstorm

What to build

• Create src/hooks/src/task-created/ directory with unified-dispatcher.ts
• Handler 1: task-metadata-injector — enriches task with project context (branch, milestone)
• Handler 2: task-dependency-validator — checks blockedBy tasks exist and are valid
• Handler 3: task-progress-tracker — logs task creation to .claude/logs/task-activity.jsonl

Pattern to follow

// Mirror TaskCompleted pattern:
// src/hooks/src/task-completed/completion-tracker.ts
// src/hooks/src/task-completed/task-commit-linker.ts
// src/hooks/src/task-completed/task-progress-tracker.ts

hooks.json registration

{
  "event": "TaskCreated",
  "command": "${CLAUDE_PLUGIN_ROOT}/hooks/bin/run-hook.mjs task-created/unified-dispatcher",
  "description": "Track task creation, inject metadata, validate dependencies"
}

Top candidates with recommended paths

# implement/SKILL.md
paths:
  - "src/**/*.{ts,tsx}"
  - "package.json"
  - "tsconfig.json"
  - "CLAUDE.md"

# design-to-code/SKILL.md
paths:
  - "src/components/**/*.{tsx,css}"
  - "**/*design-tokens*"
  - ".storybook/**"

# memory/SKILL.md
paths:
  - ".claude/memory/**"
  - ".claude/projects/**/memory/**"

# storybook-mcp-integration/SKILL.md
paths:
  - ".storybook/**"
  - "src/**/*.stories.{ts,tsx}"

# database-patterns/SKILL.md
paths:
  - "**/migrations/**"
  - "**/models/**"
  - "alembic.ini"

Full list of candidates

• implement — src files, package.json, configs
• design-to-code — components, design tokens, storybook
• memory — .claude/memory directory
• storybook-mcp-integration — storybook config, stories
• mcp-patterns — .mcp.json, MCP server files
• database-patterns — migrations, models, alembic
• cover — test files, coverage config
• fix-issue — src files for debugging
• devops-deployment — .github/workflows, Dockerfile, k8s
• ci-cd-engineer — CI config files
• security-patterns — auth files, middleware
• architecture-patterns — src directory structure
• next-best-practices — app/, pages/, next.config
• terraform-style-guide — *.tf files
• configure — .claude/settings*, plugin manifests

Where to add structured output

• brainstorm — Phase 2 divergent ideas: schema for { ideas: [{title, description, feasibility}] }
• implement — Architecture plan: schema for { components: [{file, purpose, dependencies}] }
• cover — Coverage matrix: schema for { coverage: [{file, testFile, coveragePercent}] }
• verify — Verification results: schema for { checks: [{name, pass, details}] }
• assess — Rating output: schema for { score, dimensions: [{name, score, reason}] }

Implementation pattern

// In Agent() calls within workflow skills:
Agent(
  subagent_type="backend-system-architect",
  prompt="Analyze architecture for...",
  // NEW: structured output schema
  jsonSchema={
    type: "object",
    properties: {
      components: { type: "array", items: {...} },
      risks: { type: "array", items: {...} },
      recommendation: { type: "string" }
    }
  }
)

Note: This is a documentation/skill-body change — Agent() calls are in SKILL.md prose, not compiled code.

Features to add

| Feature | Min CC Version | OrchestKit Usage |
|---------|---------------|------------------|
| TaskCreated hook event | 2.1.84 | task-created/ handlers |
| WorktreeCreate type: "http" | 2.1.84 | worktree-lifecycle-logger |
| paths: YAML glob list (skills) | 2.1.84 | 15+ skills |
| ANTHROPIC_DEFAULT_*_MODEL_SUPPORTS | 2.1.84 | ork.settings.json |
| ANTHROPIC_DEFAULT_*_MODEL_NAME | 2.1.84 | /model picker |
| CLAUDE_STREAM_IDLE_TIMEOUT_MS | 2.1.84 | Long-running agents |
| allowedChannelPlugins | 2.1.84 | Enterprise template |
| x-client-request-id header | 2.1.84 | Debug timeouts |
| PowerShell tool (Windows) | 2.1.84 | Not yet |
| MCP 2KB description cap | 2.1.84 | Awareness |
| MCP server deduplication | 2.1.84 | Automatic |
| Workflow subagent json-schema fix | 2.1.84 | bare-eval, workflows |
| System-prompt caching + ToolSearch | 2.1.84 | Automatic |
| Idle-return 75min prompt | 2.1.84 | Automatic |
| owner/repo#123 link format | 2.1.84 | Skill output |

Current handler

worktree-lifecycle-logger.ts handles WorktreeCreate with monorepo detection and sparsePaths advisory. Needs update to:

• Detect type: "http" in hook input
• Return hookSpecificOutput.worktreePath in response JSON
• Enable remote/API-triggered worktree creation for CI/CD and remote agents

// Updated handler pattern:
if (input.type === 'http') {
  return {
    hookSpecificOutput: {
      worktreePath: computedWorktreePath
    },
    // ... existing monorepo detection
  };
}

Settings to add

// ork.settings.json env section (commented out, for 3P users):
"env": {
  // ... existing vars ...
  // "ANTHROPIC_DEFAULT_OPUS_MODEL_SUPPORTS": "effort,thinking",
  // "ANTHROPIC_DEFAULT_SONNET_MODEL_SUPPORTS": "effort,thinking",
  // "ANTHROPIC_DEFAULT_HAIKU_MODEL_SUPPORTS": "effort",
  // "ANTHROPIC_DEFAULT_OPUS_MODEL_NAME": "Custom Opus",
  // "ANTHROPIC_DEFAULT_OPUS_MODEL_DESCRIPTION": "My custom Opus deployment"
}

Documentation updates

• Setup skill: add 3P provider configuration section
• Configure skill: document env var options
• Doctor skill: add to version matrix

// ork.settings.json
"env": {
  // ... existing vars ...
  "CLAUDE_STREAM_IDLE_TIMEOUT_MS": "180000"  // 3 min for long agents
}

Skills to update

• create-pr — output references like "Created PR owner/repo#123"
• fix-issue — "Investigating owner/repo#456"
• github-operations — all issue/PR references in examples
• commit — "Closes owner/repo#789" in commit messages
• review-pr — PR review output references
• visualize-plan — issue reference inputs

Pattern

# Before (CC < 2.1.84):
"Created PR #123"

# After (CC >= 2.1.84):
"Created PR owner/repo#123"  // clickable link

# In code, use git remote to detect owner/repo:
git remote get-url origin | sed 's/.*github.com[:/]\(.*\)\.git/\1/'

• #17 System-prompt caching + ToolSearch — OrchestKit uses ENABLE_TOOL_SEARCH=auto:5. Caching now works with ToolSearch enabled. Free perf boost.
• #18 MCP startup parallelism — REPL renders before all 7 MCP servers connect. Faster startup for OrchestKit users.
• #19 MCP cache leak fix — Fixed on server reconnect. 7 MCP servers benefit from no memory leak.
• #20 Startup 30ms improvement — setup() runs in parallel with slash command loading. Small but free.
• #21 p90 cache rate improvement — Better prompt caching. Token savings for all OrchestKit sessions.