Spike M9-A0: validate pinned OpenClaw 2026.4.14 hooks and bootstrap

[shenjiecode]

Scope

Run the standalone M9-A0 research spike against the pinned OpenClaw 2026.4.14 contract before any M9-A implementation begins.

This issue produces integration evidence and an explicit hook/bootstrap integration decision. It does not enable the M9-A product contract, implement trusted execution, or authorize an OpenClaw upgrade.

Required serial order

The next layer may start only after the previous layer passes:

1. Source/type/contract inspection
   • Inspect the pinned source and types for before_tool_call, tool_result_persist, before_prompt_build, bootstrap/system context, and observation hooks.
   • Record call sites, await/block behavior, throw/timeout behavior, priority, and error propagation.
2. Container hook runner + fake provider
   • Run the real hook runner in the actual build image with no external network.
   • Capture the final fake-provider request and exercise block, persistence ordering, replay, truncation, bootstrap placement, and corrupt/missing bootstrap behavior.
3. Deterministic plugin/gateway/session integration
   • Use the real Tiangong plugin and OpenClaw gateway/session path with fake provider, fake coordination endpoint, synthetic Task and Work-Leader admission, runtime bindings/claim fixtures, and a disposable workspace.
   • Verify only the bounded Task and Leader hook/bootstrap/spool/result-or-typed-action contracts.
4. Basic Matrix turn
   • Use the official Channel Plane, one reserved safe fixture, and one real member turn.
   • Confirm only that the Matrix route traverses the already-proven hook/bootstrap path; do not use it to prove authorization, concurrency, atomicity, or recovery.

Stop conditions

• Any layer failure stops the spike and all later layers.
• Do not silently upgrade OpenClaw, add a workaround, or start M9-A implementation while a pinned contract is uncertain.
• Keep credentials, tokens, private session data, and unrestricted transcripts out of committed artifacts.

Deliverables

• Sanitized plan.md before execution.
• Sanitized result.md with per-layer pass/fail, direct machine facts, limitations, and the integration decision.
• Direct bounded machine evidence for each completed layer, with exact pinned version and fixture/resource ownership.
• If a premise fails, a follow-up design revision issue/PR before implementation.

References

• M9 design PR: docs: add accepted M9 professional runtime design #109 (merged)
• M9-A0 contract: docs/design/m9-professional-agent-runtime-and-project-knowledge.md, §§16.1, 17.1, 20.1

[shenjiecode]

M9-A0 executed strictly serially and stopped at layer 1 (source/type/contract inspection).

• Result PR: test: record M9-A0 pinned OpenClaw contract stop #111
• Follow-up design issue: Resolve M9-A0 pinned OpenClaw hook and bootstrap contract blockers #112
• Layers 2–4 were not started.
• No M9-A implementation, OpenClaw upgrade, Matrix fixture, or external test resource was used.

The direct evidence shows fail-open before_prompt_build, ordinary/missing/truncated bootstrap context without digest validation, synchronous fail-open tool_result_persist, local-only current capture, and no hook-level before_tool_call timeout contract.