fix(memory): normalize OpenClaw content blocks + BM25 OR fallback + audit_log sanitization

[zhangtaibo]

Fix: agent-memory auto-capture trigger + BM25 OR fallback + audit_log sanitization

Fixes #1405

Problem

Three issues in agent-memory:

1. OpenClaw adapter auto-capture (openclaw/src/index.ts): agent_end hook used lastAsst.content directly as a string, but OpenClaw sends content as an array of content blocks [{type:"text", text:"..."}]. This caused re.test(lastAsst.content) to coerce the array to "[object Object]", so no trigger regex ever matched and auto-capture never fired.

2. BM25 FTS5 AND-only query (src/index/store.rs): When any single query term is absent from the corpus, the implicit-AND FTS5 MATCH returns 0 results. Natural-language prompts routinely contain common words not in any memory file, causing every AND query to fail silently.

3. audit_log query leakage (src/tools/memory_search.rs): format!("{:.120}", query) wrote the first 120 chars of user queries into the audit log path field, potentially leaking sensitive information.

Fix

1. Normalize lastAsst.content from content-block array to string before trigger matching and hashing.
2. Add OR-joined fallback: when AND returns 0 rows and there are multiple tokens, retry with '"token1" OR "token2" OR ...' so partial matches still surface.
3. Replace format!("{:.120}", query) with format!("bm25:len={}", query.len()) across all audit_log calls.

Verification

• Fresh clone + git apply patch + bash scripts/rpm-build.sh agent-memory → exit 0
• RPM: agent-memory-0.2.1-1.alnx4.x86_64.rpm (5.0M)
• Tests: test_auto_capture_persists_trigger_phrase, test_auto_recall_injects_without_tool_call, test_memory_get_reads_seeded_file (to be verified on ECS with installed RPM)

Test Cases

• tests/test_openclaw_e2e.py::TestAutoCapture::test_auto_capture_persists_trigger_phrase
• tests/test_openclaw_e2e.py::TestAutoRecall::test_auto_recall_injects_without_tool_call
• tests/test_openclaw_e2e.py::TestMemoryGet::test_memory_get_reads_seeded_file

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

云孟 seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}