test(scenarios): reproducible test scenarios harness

[NagyVikt]

Summary

Adds tests/scenarios/ — a reproducible, in-process scenarios harness for canonical multi-agent situations, with deterministic clocks and diffable plaintext expected-state files.

• 5 canonical scenarios as seed.sql + inputs.jsonl + expected.json
  1. 01-claim-before-edit — Codex pre_tool_use auto-claims target
  2. 02-cross-runtime-handoff — Codex relays out, Claude accepts, claim flips
  3. 03-stale-claim-sweep — TTL expiry triggers release_expired_quota
  4. 04-plan-claim-adoption — Queen sub-task adopted by Codex
  5. 05-path-mismatch-reclaim — Pre_tool_use re-claims after mismatch
• Harness uses vi.useFakeTimers() + vi.setSystemTime(BASE_TS + at_ms) + path normalization
• 4 scripts: pnpm scenarios, scenarios:filter, scenarios:explain, scenarios:record
• 2 harness self-tests (fails-closed on missing expected, clear diff on mismatch)
• Separate scenarios CI job on Node 20

Closes ⏳ Reproducible test fixture set under tests/scenarios/ under README §v0.x "Multi-runtime confidence".

OpenSpec

openspec/changes/scenarios-harness-2026-05-16/CHANGE.md

Test plan

• pnpm scenarios — 7/7 pass (5 scenarios + 2 self-tests, 1.95s)
• pnpm scenarios:filter 03-stale-claim-sweep — 1 pass, 6 skip
• pnpm scenarios:explain 02-cross-runtime-handoff — prints timeline
• pnpm scenarios:record <slug> — confirmed regenerates expected.json
• pnpm build — clean

Notes

• Uses .mts for harness internals (tsx CJS-fallback workaround for @colony/compress ESM-only exports)
• Adds task envelope kind beyond the design's lifecycle|mcp|tick triplet to drive TaskThread directly
• Pre-flips the v0.x README line ⏳ → ✅ — revert if you prefer to land that separately

Merge order

Independent of #573 (bridge replay). Safe to merge second.

🤖 Generated with Claude Code