scout — Real-System Best Practices & Production Archetype Engine

Domain Rubric

1. Source Authority Hierarchy

• Tier 1 (Canonical Standards & Official Specs): RFCs, language specifications, cloud vendor reference architectures, and formal technical papers.
• Tier 2 (High-Scale Production Engineering): Tech blogs and architecture disclosures from proven high-scale engineering organizations (e.g., Stripe, Netflix, Cloudflare, Discord, Uber, Meta).
• Tier 3 (Battle Scars & Postmortems): Documented production outages, postmortems, and root-cause analyses revealing edge cases, operational bottlenecks, and failure modes.
• Exclusion Filter: Exclude SEO-optimized content farms, unverified personal blog posts, vendor marketing collateral, and unvetted AI summaries.

2. Search & Inclusion Budget

• Exploration Ceiling: Maximum 8 search queries and 10 fetched pages per scouting run.
• Inclusion Budget: Synthesize from 3 to 5 premier, high-signal sources.
• Short-Circuit Stop: Terminate search as soon as the canonical pattern and known failure modes are triangulated with primary evidence.

Canonical Output Contract

# Scout Report: <Topic / Inquiry>

## 1. Executive Summary
- **Canonical Production Archetype**: <Boring, battle-tested standard approach adopted by industry leaders>
- **Operational Reality**: <Key complexity, maintenance overhead, or scaling bottleneck in practice>

## 2. Industry Precedents & Battle Scars
| Source / Organization | Pattern Deployed | Production Scale / Context | Known Failure Mode / Trade-off |
| :--- | :--- | :--- | :--- |
| <Org / Paper> | <Architecture / Tool> | <Traffic, Data Volume, SLA> | <Failure point, latency spike, cost trap> |

## 3. Recommended Baseline vs. Traps
- **Recommended Baseline**: <Simplest robust architecture meeting requirements without premature optimization>
- **Premature Traps to Avoid**: <Over-engineered patterns or hype-driven tools that introduce unnecessary failure modes>

## 4. Primary Citations
- [<Source Title>](<Verified URL>) — <Core insight or concrete data point extracted>

Routing

• Deliberate trade-offs or evaluate architectural fit for the project → (via /clarify).
• Baseline is accepted, adequately supported, and the caller requests change planning → (via /changeset).

Execution Protocol

1. Query Triangulation & Dispatch:
   • Formulate 2 to 3 targeted search queries crossing: canonical standards/RFCs, high-scale engineering cases, and failure postmortems.
   • Launch

     research

     subagent(s) dynamically scaled to inquiry complexity or caller-declared budget (Role:

     Real-System Research Scout

     ) supplying:
     • The triangulated queries and Search Budget (max 8 queries, 10 fetched pages).
     • The Canonical Output Contract above as the required output contract (offload deep comparative dossiers to

       .agents/scratch/scout-dossier.md

       when extensive).
   • Mandate the subagent to inspect full page bodies via URL tools before synthesizing; cite strictly verified URLs and never synthesize from search snippets alone.
2. Synthesis & Halt:
   • Emit the completed report into the conversation stream (with pointers to any offloaded dossiers) and halt turn immediately.