14 CCA-F scenarios: 6 official + 8 production deep-dives.

A production refund/escalation agent built on Claude. The harness reads stop_reason, dispatches tools through a registry, gates risky operations with a PreToolUse hook (deterministic policy

A multi-turn conversational pattern that survives context compression. The harness pins a CASE_FACTS block at the top of every system-prompt iteration (immutable, re-read every turn), summar

Claude Code as a team-aware code-generation agent. The repo carries a CLAUDE.md hierarchy (project + personal + system) committed to version control, area-specific rules in .claude/rules/ ke

A hub-and-spoke research system. The coordinator owns task decomposition (semantic, not lexical), spawns 3-5 research subagents in parallel with isolated contexts and scoped tools, routes fi

Claude Code as a headless PR reviewer in GitHub Actions. The workflow uses claude -p for non-interactive execution, runs per-file independent sessions (no shared context across the 14 files

An enterprise-scale Skills registry. Each Skill is a markdown file with frontmatter (name + version + description + tags + dependencies + access_level), stored in .claude/skills/{team}/{name

A team of specialised subagents that handle codebase exploration, code review, and doc generation. Each with a narrow tool whitelist and isolated context. The lead agent uses Grep + Glob fir

A schema-driven extraction agent. The harness defines the output shape as a tool with input_schema, sets tool_choice to force that tool, runs a validation-retry loop (parse → validate → on f

A meta-skill scenario about designing tool registries. The optimum is 4-5 tools per agent (past 5, routing accuracy drops 8% per extra tool); each tool follows the Anthropic 4-line descripti

A retrieval-augmented long-document agent. Semantic chunking (paragraph- and section-aware, not fixed-size) preserves meaning; an embedding index supports top-K retrieval so only the relevan

An on-call ops agent. Alerts arrive structured (service, severity, root_cause_signal, 4-bucket isError: Transient · Permission · Data · Business); the agent matches a runbook from a small re

A CLI-first Skills architecture for developer tooling. Each Skill is a markdown file with frontmatter (name, version, parameters, allowed-tools); the CLI invokes them via claude skills invok

A code-execution Skill with four layers of safety. Layer 1: route file I/O to built-in tools (Read, Write, Edit) and reserve Bash only for actual execution. Layer 2: a PreToolUse hook scans

An AP-automation agent that wraps four guarantees around invoice approval. (1) Forced tool_use with a strict JSON schema (vendor_id, invoice_number, line_items[], total_amount, currency ISO