[WIP] Downgrade Agent Persona Explorer to claude-haiku-4-5 and optimize data fetching

[Copilot]

Thanks for asking me to work on this. I will get started on it and keep this PR's description up to date as I form a plan and make progress.

---

This section details on the original issue you should resolve

<issue_title>[deep-report] Downgrade Agent Persona Explorer to claude-haiku-4-5 and move data-fetching to pre-agent shell steps</issue_title>
<issue_description>### Description
Agent Persona Explorer runs 42 turns, consumes 2.2M tokens, and produces only 1 safe output — the observability kit flags it resource_heavy + poor_agentic_control + partially_reducible (95% data-gathering turns) + model_downgrade. 95% of the agent's work is fetching and reading persona data that could be retrieved deterministically before the agent starts. The current Opus/Sonnet model is massively over-provisioned for a task that boils down to "generate 4 workflow templates from pre-fetched persona descriptions."

Expected Impact

Moving persona data retrieval (agent definitions, workflow examples) to pre-agent shell steps would reduce turns from 42 to ~8. Downgrading to claude-haiku-4-5 would cut per-run cost by ~80% while maintaining output quality (the task is template generation, not complex reasoning). Estimated savings: ~$1.80/run based on current token volumes.

Suggested Agent

Any agent with access to .github/workflows/agent-persona-explorer.md. Add pre-agent-steps to read and concatenate persona files, then pass as context to a simplified prompt with max_turns: 10.

Estimated Effort

Medium (1–4 hours) — requires pre-agent step scripting + prompt redesign.

Data Source

DeepReport Intelligence Briefing — 2026-04-21 (run §24729745351); corroborated by Agent Performance Report #27491, Observability Report #27292.

Generated by DeepReport - Intelligence Gathering Agent · ● 586.9K · ◷

• expires on Apr 23, 2026, 3:15 PM UTC

Comments on the Issue (you are @copilot in this section)