Skip to content

Transmute-Data fit

Status: Draft for discussion. Anchors abstract runtime design to repositories and conventions in the parent Transmute-Data tree without turning this repo into an implementation guide.

Workspace context

ai-system-design is a sibling documentation project under /home/jason/Transmute-Data, alongside agent platform repos (stn-agents, stn-agent-eval, stn-coder, transmute-agent-tools, transmute-agent-skills) and Tayne-oriented trees under tayne_projects/. Secrets stay in agent_config/ or environment variables, not in this documentation repo.

Eval harness (stn-agent-eval)

JSONL suites and stn-agent-eval run provide repeatable scenarios for harness behavior. Candidate mappings:

  • Harness quality: tool selection, task completion, and regression detection after prompt or skill changes.
  • Runtime contract tests (future): scenarios that require durable resume, streaming, or HITL, only where the harness under test is wired to a runtime that exposes those primitives.

Open question: whether production-runtime SLO tests live in the same harness or a separate suite in tayne_projects/transmute-agent-tools-eval.

Operator CLI (transmute-agent-tools, td)

The td CLI integrates Grafana, Jira, Slack, and related operator workflows with token discovery documented in transmute-agent-tools/docs/STN_AGENT_CONFIG.md. For agent system design, treat td as the pattern for human operator tools adjacent to the agent runtime, not as a substitute for tenant-facing agent APIs.

Coder workspaces (stn-coder)

Coder deployment and workspace bootstrap (including Claude Code and Superpowers scripts) define where harness development and eval runs happen for STN. Runtime deployment targets may differ from Coder dev workspaces; document assumptions when a design depends on long-lived local processes versus managed Agent Server infrastructure.

Architecture and remediations

stn-agents/docs/ and the DT matrix are the canonical STN architecture references. Remediation workflows that set STN_REPO_ROOT to a parent containing remediations/ are operational context for agents that change infrastructure; link harness capabilities to those workflows when designing SRE-facing agents.

Tayne two-repo layout

tayne_projects/transmute-agent-platform and tayne_projects/transmute-agent-tools-eval carry their own docs/WORKPLAN.md milestones. This design doc set should cross-link when Tayne runtime choices diverge from STN patterns rather than duplicating workplans here.

Suggested attachment points (not yet written)

  • Matrix from each runtime capability row to existing STN or Tayne components (if any).
  • Example eval JSONL scenarios per capability once runtime endpoints exist.
  • Coder workspace checklist for authors editing harness versus operators observing production.