Skip to content
/
Phase D Batch 6

Batch 06 — Claude Code Playbooks & Guides

Batch 06 — Claude Code Playbooks & Guides

Roster (10)

slug stars distribution cli_binary local_ui orchestration multi_model tier
claude-code-playbook-toy-crane 3 claude-plugin no yes (Next.js lecture site) sequential yes (opus+sonnet) A
claude-code-playbook-riya 26 methodology-doc no no none no B
claude-code-playbook-avinash 2 methodology-doc no no none no C
claude-code-ultimate-guide 4500 methodology-doc + MCP no no sequential no A
claude-code-templates-davila 27587 npm-package yes (installer) yes (aitmpl.com) parallel-fan-out yes (per-agent) A
claude-code-system-prompts 10516 methodology-doc no no none no A
ai-dev-workflow-tkhongsap 1 standalone-repo no no hierarchical no B
ai-dev-workflow-kit-aahil 4 standalone-repo no no sequential no B
ai-task-master-james 21 npm-package yes (task-master) no sequential yes (OpenRouter) B
learn-claude-code 62800 methodology-doc no no none no A

Intra-batch patterns

Namesake cluster: Three repos share the name "claude-code-playbook" (toy-crane, RiyaParikh0112, avinashmyerolkar). Each is entirely independent — different authors, different purposes, different architectures. toy-crane is a live Korean lecture site + plugin; RiyaParikh0112 is an English guidebook with no deployed tooling; avinashmyerolkar is a personal study notebook. The name collision is coincidental community convergence, not a fork relationship.

Slash command surface divergence: Among the ten frameworks, slash-command surface varies from 0 (RiyaParikh0112, avinashmyerolkar, ai-dev-workflow-kit-aahil, learn-claude-code) to 9 (claude-code-ultimate-guide) to ~16 CLI subcommands (ai-task-master-james, davila7). The frameworks with the richest slash-command surfaces are also those with the most active maintenance and highest star counts.

Opinionation spectrum: The batch spans from highly opinionated (ai-dev-workflow-kit-aahil with Max-2 Rule, EPCC engine, Decision Gates; ai-dev-workflow-tkhongsap with mandatory clarifying questions) to completely unopinionated (davila7 — pure toolbox, no workflow structure; RiyaParikh0112 — advisory only). The two most opinionated frameworks have the fewest stars (<5); the least opinionated framework (davila7) has 27,587 stars — suggesting opinionation actively limits adoption.

Educational vs deployable split: 6 of 10 frameworks are primarily educational (methodology-doc distribution type). Only 3 ship deployable npm packages (davila7, ai-task-master-james) or a plugin (toy-crane). learn-claude-code is the standout: 62,800 stars as a pure curriculum with no deployable artifacts.

Security consciousness: claude-code-ultimate-guide is notable for the 28-CVE threat database and dangerous-actions-blocker.sh hook. claude-code-system-prompts provides the forensic foundation for security research by exposing all internal CC prompts. No other framework in the batch treats security as a first-class concern.

Most interesting finds

  1. claude-code-system-prompts (Piebald AI): The only system-prompt observatory in the entire corpus. Reveals that Anthropic's internal Claude Code already implements an 8-stage multi-agent code review pipeline, dream memory consolidation/pruning, and security monitoring subagents — capabilities most community frameworks are unknowingly re-implementing from scratch. The companion tweakcc product for surgical prompt patching has no equivalent in any batch.

  2. learn-claude-code (shareAI-lab): At 62,800 stars — the highest-star framework in this entire batch by 2.3x — yet ships zero deployable primitives. Its philosophical framing ("agency is trained, not coded; harness engineers build vehicles") is the most rigorous intellectual anti-pattern argument in the corpus. The hook teaching principle "hang on the loop, don't write into it" is the clearest pedagogical statement of the extension-without-modification pattern.

Items written as Tier C stubs

None. All 10 frameworks had sufficient public material for full 11-file reports.

Note: claude-code-playbook-avinash (2 stars, 50% incomplete) was close to Tier C but had enough documented content (SDD notes, slash command taxonomy) for meaningful analysis.

Cross-references discovered

  • ai-task-master-james → forks taskmaster-ai seed (eyaltoledano/claude-task-master). canonical: false.
  • claude-code-playbook-toy-crane → ships /migrate-to-copilot skill for converting FROM the toy-crane Claude Code plugin format TO Copilot CLI format — acknowledging platform portability as a first-class concern.
  • claude-code-ultimate-guide explicitly self-positions against everything-claude-code (another community reference in the broader corpus, not in this batch).
  • avinashmyerolkar and ai-dev-workflow-tkhongsap both articulate near-identical Vibe Coding vs SDD comparison tables — suggesting these emerged from the same community discussion thread or shared source.
  • claude-code-system-prompts documents the skill-generate-permission-allowlist-from-transcripts skill, which is architecturally similar to the fewer-permission-prompts superpowers skill — Anthropic and the superpowers community solved the same problem independently.