Codex DM mid-session turns should not rediscover campaign state via shell scans

[100yenadmin]

Summary

The 2026-06-01 current-head built-app Codex-DM short playtest on e79144d stayed playable, but both post-opening turns spent avoidable time discovering live campaign state through shell/file scans before using engine MCP tools.

Evidence

• Evidence folder: /Volumes/LEXAR/Codex/worldos-built-app-playtest/codex-app-nullfix-20260601T032041/
• Built app run: codex-app-nullfix-20260601T032041
• Minted run: play-20260531202048
• Provider trace summary: 07-provider-trace-summary.json
• Trace showed command_execution_count: 32, including rg, find, and direct chat.jsonl/player_moves.jsonl reads during mid-session turns.

User Impact

The app accepts the player move immediately, but the DM response latency is inflated by avoidable state discovery. This makes a fresh-player turn feel stalled even though the engine is working.

Acceptance Criteria

• The wrapper passes the live campaign id into mid-session prompts whenever it can discover one from the run snapshot.
• Mid-session prompts instruct Codex to use scene_context / engine MCP tools, not shell/rg/find/filesystem scans, for live campaign state.
• A repeat built-app move proof shows the turn can resolve through engine tools without broad repo/state discovery.

[100yenadmin]

Addressed on PR #475 at commit 8031445 and carried through final head 500c379. The final exact-head provider trace summary at /Volumes/LEXAR/Codex/worldos-built-app-playtest/codex-app-final-20260601T033714/05-provider-trace-summary.json shows the post-move turn went straight from the expected DM skill read to scene_context(campaign_id="camp_39109282f754"); no broad rg, find, or state-file discovery after the player move. Leaving open until #475 merges.

[100yenadmin]

Closed after PR #475 merged. The final PR #475 traces show mid-session turns using scene_context(campaign_id=...) instead of broad rg/find state rediscovery. Any remaining provider trace-noise is tracked in #479.