Memory hooks: a small schema gap, and an offer to help with a Guardian for them

[vgudur-dev]

Hi all, and thank you for ACS. I work on agent memory security (I lead OWASP Agent Memory Guard), so I read the two memory hooks closely. Two things:

1. A small spec point. docs/spec/instrument/hooks.md lists TTL in the memoryStore payload, but specification/v0.1.0/hooks/memory-store.json has no TTL or expiry property, only free-form metadata. Would you like an issue or a small PR to align them? An explicit expiry field would help, because it lets a Guardian limit how long an entry may live.

2. An offer. The AGT reference implementation evaluates steps/toolCallRequest and steps/toolCallResult. I'd be glad to contribute a small Python Guardian that evaluates steps/memoryStore and steps/memoryContextRetrieval, using Agent Memory Guard as the policy engine, with a few conformance vectors for those two hooks. Examples: a write that looks harmless but changes later behaviour, and a stored entry that should be denied when it is read back.

This sits next to #507 (memory lineage), which covers provenance; I'd follow whatever that settles.

I know the current priority is the runnable Guardian and the interoperability evidence, so please tell me whether this fits now or should wait for v0.2.0. Happy to follow whatever process you prefer.