Guard Memory Writes Against Injection

PROJECT CONTEXT:
My agent writes to long-term memory, and anything it reads could plant a false 'fact' that persists across sessions - memory poisoning. Memory-validation research treats every memory write as untrusted input needing validation, provenance, and re-verification.

GOAL:
Harden the memory write path against poisoning: source-aware write policy, content validation, provenance tracking, and scheduled re-validation.

REQUIREMENTS:
- Write policy keyed on the trust of the content's source
- Validation of memory content before persistence (format, plausibility, conflict)
- Provenance stamped on every memory: source, trust level, session, timestamp
- Quarantine for writes derived from untrusted content until corroborated
- Scheduled re-validation that expires or flags stale/contradicted memories

CONSTRAINTS:
- Untrusted-sourced content can never write high-trust memory directly
- A memory's influence must be traceable to its provenance at recall time

DEFINITION OF DONE:
- A 'fact' derived from a fetched web page lands in quarantine, not core memory
- Every stored memory shows source and trust on inspection
- A poisoning attempt via injected content fails to become an acted-on belief
- Re-validation expires an unconfirmed quarantined memory after its TTL

COMMON FAILURES TO AVOID:
- Persisting anything the agent 'learned' without asking where it came from
- Memories with no provenance, so a poisoned fact is indistinguishable from a real one
- Untrusted-sourced claims recalled with full authority
- Never re-checking stored facts, so poison persists indefinitely