Run the three-probe empirical verification of Task-dispatched hook coverage

[ivklgn]

Problem

hooks-validation-system.spec.md documents that the Write|Edit PreToolUse hooks and the MCP PostToolUse hooks fire for writes performed by Task-dispatched sub-agents, but the live empirical probe that would confirm it was deferred and never captured with a dated result. The guarantee is asserted, not verified.

Expected result

Run the documented three-probe protocol in a fresh claude --plugin-dir <repo> session — Task(general-purpose) → Write src/probe.ts and → Write .archcore/probe.adr.md — confirm check-code-alignment and validate-archcore fire, and record the dated outcome in the spec. If the sub-agent stdin shape differs from a main-session call, extend bin/lib/normalize-stdin.sh and pin it with a bats case.

Impact

Either confirms the delegated-agent coverage guarantee with evidence, or surfaces a real gap — in which case it becomes the trigger to revisit sub-agent snapshot injection. Low effort, high assurance.

Proposed approach

• Run the probe protocol; record the dated result in the spec/commit.
• Only if stdin differs: extend normalize-stdin.sh + add a bats case.

Acceptance criteria

• A dated empirical result is recorded in hooks-validation-system.spec.md.

References

Design rationale: .archcore/plugin/jtbd1-phase2-hardening-delegated.plan.md (remaining follow-up). Verification chore — the spec forbids committing probe lines into bin/.