Spike: audit tech-lead agent scope — keep, decompose, or extract?

[cpliakas]

Driving Question

Is the current scope of agents/tech-lead.md (routing + orchestration tiers + convention domain ownership + model override governance) appropriate for one agent, or should some responsibilities extract to a Router specialist, a dedicated agent, or a skill?

Why Now

Code churn analysis for the 30-day window ending 2026-04-24 flagged agents/tech-lead.md as the #2 hotspot and a thrashing file:

• Grew from 457 → 782 lines (71%) between 2026-03-29 and 2026-04-21
• 9 changes in the analysis window, 5 concentrated on 2026-04-21 (release-day burst of 4 parallel features)
• Co-changes with README.md in 78% of commits (tight coupling, likely indicating both files are acting as manifests rather than stable contracts)
• Co-changes with agents/qa-lead.md in 100% of qa-lead's commits (n=3, small sample)

The pattern suggests the Tech Lead is absorbing responsibility from every new feature. Before the next round of features lands, we want a deliberate decision on whether the agent's scope is still right.

Key Questions to Explore

• What top-level responsibilities does tech-lead.md currently hold? (Enumerate by section.)
• Which of those responsibilities are inherent to a tech-lead role vs. which were placed there by convenience?
• Is there a coherent sub-domain (e.g., routing) large enough to justify extracting to a specialist agent?
• Are any responsibilities better expressed as skills the Tech Lead invokes, rather than instructions it carries?
• How do sibling agent files compare in size and scope? Does Tech Lead's breadth look anomalous or proportional?

Options to Consider (not exhaustive)

Option	Summary
Keep as-is	Agent breadth reflects the role; size is a side-effect, not a defect
Extract Router specialist	Routing logic moves to a new agents/router.md under the Tech Lead
Extract to skills	Orchestration tiers and/or convention domain ownership move to skill files
Reorganize sections only	No structural change; improve internal organization of the existing file

Out of Scope

• Any changes to other agents' scope
• Churn rate improvements as a goal in themselves — the metric is a signal, not a target
• Rewriting the agent's behavior; this spike is about structure, not capability

Timebox

1–2 hours. Produce a recommendation doc even if not all questions are fully answered.

Exit Criteria

• A written recommendation: keep / decompose / extract-to-skill
• If decomposition or extraction is recommended: a follow-up proposal ticket with the split
• Route the recommendation through the Chief Architect (per the write-spike skill's guidance that architectural trade-offs need Architect review)

Related

• Code churn analysis run on 2026-04-24 (30-day window)
• Commits contributing to the growth: e4d33374, 63052b43, and the 2026-04-21 release-day merges