Skip to main content

​
What Happened

Before executing any changes, launched 3 parallel assessment agents:
  1. Service implementation agent — read every Go/TypeScript source file across 5 services
  2. Architecture/contracts agent — read all proto, OpenAPI, infrastructure, requirements files
  3. Secrets/gists agent — scanned 100 GitHub gists + ESO manifests for secret coverage
All three returned comprehensive reports simultaneously. The combined assessment revealed the SDK was 85% complete (not 40%), golden-press had a schema BLOCKER, and 3 Helm values files were missing.

​
The Prompt Pattern

# Launch 3 agents in parallel with domain-specific prompts:
Agent 1 (Explore): "Read every source file in services/. Report what's implemented vs stubbed."
Agent 2 (Explore): "Read proto/, contracts/, infrastructure/. Report completeness."  
Agent 3 (General): "Run gh gist list, filter for Choco relevancy, read contents."
Critical rule: Each agent gets a complete, self-contained brief. No agent references another agent’s work. The orchestrator (main context) synthesizes all three reports before producing the roadmap outline.

​
Why This Matters for FORGE/TOMMY

This pattern maps directly to FORGE pipeline stages:
  • DISCOVER = parallel assessment agents
  • DESIGN = orchestrator synthesizes and produces TASKSET outline
  • DEVELOP = gated execution per TASKSET
  • DEPLOY = staging reconciliation
  • DELIVER = go-live checklist
The parallel assessment phase is the highest-leverage investment. Without it, TASKSET 5 (SDK) would have rewritten 50+ methods that already existed. The 5-minute assessment saved hours of execution.

​
Reusable For

Any multi-repo platform assessment. Launch one Explore agent per domain (services, infrastructure, contracts) plus one general-purpose agent for external resources (gists, linear issues, notion pages). Synthesize before planning.