Batch 30 — Cursor/Roo Overflow + Community Packs

Roster (7)

Intra-Batch Patterns

Three of the six accessible frameworks (roomodes-sellisd, roo-code-vibook, james-cherished-roo-community) target Roo Code exclusively and have zero cross-tool portability, reflecting a community that has coalesced around Roo-specific formats (.roomodes, .roorules-*). The batch confirms that the Roo Code ecosystem has spawned distinct community tooling for mode management (james-cherished-roo-community's Visualizer), agile team simulation (roomodes-sellisd), and documentation generation (roo-code-vibook) — three different problems, all using the same underlying .roomodes format.

The cursor-agentic-pipeline framework is the batch's outlier: despite its slug name, it targets GitHub Copilot's cloud agent runtime (not Cursor), and serves a narrow Laravel PHP research use case rather than general-purpose agent workflow.

vibe-stack introduces freemium gating of rule content — the only framework in this batch or the seeds to withhold safety-critical rules behind a $29+ paywall. This is architecturally interesting because the rules it withholds (stripe-payments, file-uploads, middleware-auth) are among the highest-risk areas for AI hallucination.

cursor/plugins (cursor-rules-pack-v2) stands apart from all other batch members: it is Cursor's own first-party distribution, has 853 stars vs. the next highest's 23, ships a multi-plugin marketplace with hierarchical cloud-agent orchestration, transcript-based memory, multi-model dispatch, and the unique <promise> tag completion pattern. It is the most mature framework in the batch.

Most Interesting Find

cursor/plugins (cursor-rules-pack-v2): The ralph-loop plugin's <promise> XML tag pattern — where the AI embeds a machine-detectable completion signal in its output text, which a hook monitors programmatically — converts natural-language task completion into a verifiable condition. Combined with the pstack multi-model routing (different models for code vs. prose) and the continual-learning transcript mining system, this is the most complete "continuous improvement" architecture in the entire batch.

cursor-agentic-pipeline (secondary): The only framework targeting GitHub Copilot's .github/agents/ cloud agent format — an entirely different runtime than any seed or batch peer. Its "Surgical Output" principle (write exactly one file per invocation, nothing else) is the strictest single-invocation constraint in the corpus.

Items Written as Tier C

• roocode-memorybank-optimized — repository https://github.com/shipdocs/roocode-memorybank-optimized returned HTTP 404. Not accessible. Only 00-summary.md (status: insufficient-public-material) and METRICS.yaml (quality: tier-c-evidence-incomplete) written.

Cross-References Discovered

1. james-cherished-roo-community explicitly references and deprecates in favor of taskmaster-ai (seed) — the author's AI-task-master project is archived pointing to taskmaster-ai as the successor.
2. roomodes-sellisd and james-cherished-roo-community both target the same .roomodes format and Roo Code runtime; the sellisd version is more structured (JSON persona files + installer), the James Cherished version is simpler (curl install + web UI).
3. cursor-rules-pack-v2 (pstack) references cursor-team-kit's deslop skill across plugin boundaries — inter-plugin dependency within the same repo.
4. roo-code-vibook and roomodes-sellisd both ship Roo Code custom modes: vibook ships one mode (prompt-engineer), sellisd ships five agile team personas.
5. cursor-agentic-pipeline slug is named cursor-agentic-pipeline but targets GitHub Copilot, not Cursor — a naming mismatch that should be flagged for the catalog.