Distinguish explicit from implicit preferences when writing memory (v0.2.9)

[jackaldenryan]

Summary

• Adds an Explicit vs implicit preferences section to the skill so a one-off request inside a specific piece of work is no longer stored as a standing rule. Real case: asking to remove the em dashes from one email draft produced the memory "prefers no em dashes in his emails/writing", which then leaks into unrelated future work.
• Implicit requests are now stored qualified to what was asked and the situation it happened in, and are promoted to a general preference only when the user states it as a standing rule or repeats it across enough separate situations.
• Retrieval side matches: explicit preferences apply by default, implicit ones apply only when the current work resembles the scenario they were recorded in, and the conflict-resolution order now ranks explicit above implicit.
• Version synced to 0.2.9 across the three manifests, with a CHANGELOG entry.

Test plan

• claude plugin validate plugins/zep-memory/.claude-plugin/plugin.json --strict
• claude plugin validate . --strict
• python3 scripts/plugin_manifests.py --check
• python3 scripts/validate_agent_plugin.py
• .github/workflows/test-plugin.yml passes

Made with Cursor