Context
M9-A0 issue #110 stopped at the first serial layer. PR #111 contains the sanitized plan, result, and direct source evidence.
Blocking pinned-contract findings
Against OpenClaw 2026.4.14:
before_prompt_build failures are caught and the model path continues;
- missing bootstrap files become prompt text and oversized files may be truncated;
- the inspected bootstrap path has no digest/checksum validation;
tool_result_persist is synchronous-only and defaults to fail-open handling;
- the current Tiangong capture is a local synchronous append, not proven control-domain spool closure;
- the inspected
before_tool_call runner has no hook-level timeout enforcement.
Required decision
Revise the M9-A0 integration contract, or prove a Tiangong-owned trusted wrapper/control boundary that satisfies the design without relying on fail-open prompt hooks or synchronous observation hooks. Preserve the distinction between model/provider claims, hook outcomes, persistent records, and Matrix observations.
Exit criteria
- design revision is reviewed before rerunning the spike;
- no OpenClaw version upgrade is introduced silently;
- the rerun again follows source/type/contract → container fake-provider → deterministic integration → Basic Matrix strictly in order;
- M9-A implementation remains blocked until all four layers pass.
Refs #110
Context
M9-A0 issue #110 stopped at the first serial layer. PR #111 contains the sanitized plan, result, and direct source evidence.
Blocking pinned-contract findings
Against OpenClaw
2026.4.14:before_prompt_buildfailures are caught and the model path continues;tool_result_persistis synchronous-only and defaults to fail-open handling;before_tool_callrunner has no hook-level timeout enforcement.Required decision
Revise the M9-A0 integration contract, or prove a Tiangong-owned trusted wrapper/control boundary that satisfies the design without relying on fail-open prompt hooks or synchronous observation hooks. Preserve the distinction between model/provider claims, hook outcomes, persistent records, and Matrix observations.
Exit criteria
Refs #110