feat(memory): bundle #533 + #549 — facet guard, length limits, concise extraction

[ZaynJarvis]

Cherry-picked bundle of two in-flight prompt-quality PRs from @lishixiang0705. Original commits preserved — authorship is unchanged; my only new commit is a fixup addressing the inline review feedback.

Commits

SHA	Author	Source	Change
1c98e28	@lishixiang0705	#533	facet guard + length limits in memory_merge_bundle.yaml
2566355	@lishixiang0705	#549	length targets + vectorize-on-abstract
508576a	@ZaynJarvis	review fixup	drop BAD/GOOD blocks, restore English examples

What each piece does

#533 — memory_merge_bundle.yaml v1 → v2

• Facet coherence check: even within the same category, don't merge memories covering different facets (Python code style + food preferences stay separate). decision: skip added.
• Hard length limits: abstract ≤ 80, overview ≤ 200, content ≤ 300.
• Condensed snapshot strategy: on conflict keep newer value, do not accumulate superseded details.

#549 — memory_extraction.yaml + memory_extractor.py

• Explicit length targets on the Three-Level Structure section (~50-80 / 3-5 bullets / 2-4 sentences).
• set_vectorize(abstract or content) — shorter, keyword-dense text produces more discriminative embeddings.

Fixup — addresses inline review on #549

• Drop the BAD/GOOD content example blocks. They duplicate the Few-shot Examples section below (my comment on feat: optimize memory extraction for concise output and precise retrieval #549).
• Restore English values in the L0 bullet examples; the Chinese values would bias output_language: auto for non-Chinese users (@yangxinxin-7's point).
• Did not update the existing Few-shot Examples to match the new length targets (yangxinxin-7's first point) — separate scope, follow-up.

Memory-quality effect

Pending eval. Will run retrieval precision / embedding score-spread numbers before merge.

Not in this PR

The queue stall fix from #951 is kept standalone — see its own companion PR (unrelated code path).

Test plan

• Both YAML templates parse with yaml.safe_load
• Retrieval precision / embedding score-spread eval on a real memory corpus

[github-actions]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis ✅ 533 - Fully compliant Compliant requirements: Added facet coherence check with "skip" decision examples Added hard character limits with summarization instructions Updated to condensed snapshot strategy 549 - Fully compliant Compliant requirements: Added explicit length targets (abstract ~50-80 chars, content 2-4 sentences) Updated vectorization to use abstract or content

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🏅 Score: 85

🧪 No relevant tests

🔒 No security concerns identified

✅ No TODO sections

🔀 Multiple PR themes Sub-PR theme: feat: optimize memory extraction with concise prompts and abstract vectorization Relevant files: openviking/session/memory_extractor.py openviking/prompts/templates/compression/memory_extraction.yaml Sub-PR theme: feat: add facet guard and length limits to memory_merge_bundle Relevant files: openviking/prompts/templates/compression/memory_merge_bundle.yaml

⚡ Recommended focus areas for review Breaking Change Risk The template now returns decision: "skip", but the PR does not include updates to code that consumes this template. Callers expecting only "merge" will fail to handle the new decision value. If the two memories cover DIFFERENT facets, you MUST output decision "skip". Do NOT merge unrelated information just because they share the same category. ## Step 2: Merge (only if same facet) If merging: - When facts conflict, keep the NEWER version only. - Condense to essential facts. Do NOT accumulate every historical detail. - The merged memory should read as a clean, up-to-date snapshot — not a changelog. ## Hard length limits - `abstract`: ≤ 80 characters - `overview`: ≤ 200 characters - `content`: ≤ 300 characters If the merged result would exceed these limits, aggressively summarize. Drop older, less important details first. Preserve specific values (names, numbers, versions) over narrative. ## Output Return JSON only. Two possible decisions: When merging: { "decision": "merge", "abstract": "one-line L0 (≤80 chars)", "overview": "structured L1 (≤200 chars)", "content": "condensed L2 (≤300 chars)", "reason": "short reason" } When skipping (different facets): { "decision": "skip", "reason": "short reason why these are different facets" }

[github-actions]

PR Code Suggestions ✨

No code suggestions found for the PR.