Current Focus

I'm evolving my personal Hermes environment into a more structured multi-agent orchestration platform. The goal: a system where specialised agents can be composed declaratively for complex workflows, with built-in verification, context management, and observability.

Current Sprint

• Skill Standardisation: Converting 150+ ad-hoc skills to a unified schema with typed inputs/outputs, verification hooks, and usage analytics
• Agent Loop Engineering: Applying loop-engineering principles — systematic prompt optimisation, retry strategies, cost/latency tracking
• Context Protocol: Defining how agents pass structured context (not raw text) — JSON schemas, citation requirements, confidence scores
• Verification Pipeline: Every agent output → lint/test/build/render/cite → pass/fail → retry with error context
• Observability: Structured logging to GBrain + Grafana: token usage, latency, success rate, cost per task

Target Capability

Declare a workflow like this:

workflow "competitor-analysis":
  input: { company: "Acme Corp", depth: "comprehensive" }
  steps:
    - agent: serxcgn.recon
      task: "map digital footprint"
    - agent: youtube-channel-research.audit
      task: "analyse video strategy"
    - agent: business-opportunity-research.scan
      task: "find market gaps"
    - agent: hermes.synthesis
      task: "produce investment memo"
  verification:
    - citations_required: true
    - min_sources: 10
  output: { format: "pdf", template: "investment-memo" }

Then the orchestrator handles: delegation, context passing, verification, retries, synthesis, delivery.