You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
l3.abstraction hard-fails entire world-model generation on empty LLM title (no retry/fallback) #2335
In plugin v2.0.17, l3.abstraction throws LLM_OUTPUT_MALFORMED: 'title' must be a non-empty string whenever the abstraction LLM call returns an empty/whitespace title. This aborts the entire world-model generation for that cluster — no retry, no fallback title, no skip-and-continue — so a single malformed LLM response kills the whole run.
Location
dist/core/memory/l3/abstract.js (v2.0.17), line ~56:
thrownewMemosError(ERROR_CODES.LLM_OUTPUT_MALFORMED,"l3.abstraction: 'title' must be a non-empty string",{got: o.title});
Observed behavior (production evidence)
From api_logs in our deployment (Hermes adapter, Windows, openai_compatible LLM):
Period
Attempts
Failures
Dominant failure
2.0.14 era (Aug 10–30)
~2000
~1990
mostly host handler queue is full (separate issue)
2.0.17 (Aug 30 → Sep 3)
12
1
'title' must be a non-empty string
Sample failure event:
{"kind":"l3.failed","stage":"abstract","error":{"code":"llm_failed","message":"l3.abstraction: 'title' must be a non-empty string"},"clusterKey":"python|_"}
The feature demonstrably works when the LLM behaves (6 world models successfully created since Aug 11), but the empty-title case turns one bad sample into a full pipeline failure.
Suggested fix
Instead of throwing on empty title:
Retry once with a repair prompt ("return JSON with a non-empty title"), or
Derive a fallback title (e.g. join domainTags, or first line of the body), and only skip storing that one model — never fail the whole abstraction stage.
Note: dist/core/memory/l2/induce.js:54 has the same throw-on-malformed pattern for multiple fields; the same hardening may be worth applying there.
Summary: Fixes issue #2335: l3.abstraction no longer aborts a whole cluster's world-model generation when the LLM returns an empty title. The validate callback still requires the triple (environment / inference / constraints) but no longer throws on empty title; normaliseDraft synthesises a fallback title from domain_tags (title-cased, joined with " · "), the first non-empty environment[].label, or a static "Untitled world model". abstractDraft emits an abstract.title_fallback warn breadcrumb so ops can still spot LLM regressions.
Applied the same softening to l2.induction: title becomes a display attribute (falls back to signatureLabel, else "Untitled policy") while trigger and procedure remain load-bearing and still fail the draft when missing.
Verified with vitest: the two focused test files run 16 tests (all passing) and the full tests/unit/memory/l2 + l3 suite runs 89 tests (all passing). tsc -p tsconfig.json --noEmit is clean. Changes touch four files under apps/memos-local-plugin/{core,tests}/memory/{l2,l3}; opsp task file and openspec change artefacts are staged in .ai-tasks/ and openspec/changes/ for scheduler archival.
Base branch:main Branch:bugfix/autodev-2335-20260903031432962 Commit:c2cf01b PR:#2338 Assigned to:@whipser030 Reviewers:@whipser030, @hijzy
Activity
Memtensor-AI commentedon Sep 3, 2026
🤖 AutoDev has picked up this issue and started working on it.
Task ID:
60cd97aef73c495fWorking branch:
bugfix/autodev-2335-20260903031432962Target branch:
latest dev* branchWorkflow: opsp (analysis → coding → testing → PR)
I will post the PR link here once done. If I need more information, I will ask in the comments.
fix(l3): synthesise fallback title when abstractor returns empty (Mem…
Memtensor-AI commentedon Sep 3, 2026
✅ AutoDev task
60cd97aef73c495fcompleted.Summary: Fixes issue #2335:
l3.abstractionno longer aborts a whole cluster's world-model generation when the LLM returns an emptytitle. Thevalidatecallback still requires the triple (environment/inference/constraints) but no longer throws on emptytitle;normaliseDraftsynthesises a fallback title fromdomain_tags(title-cased, joined with " · "), the first non-emptyenvironment[].label, or a static"Untitled world model".abstractDraftemits anabstract.title_fallbackwarn breadcrumb so ops can still spot LLM regressions.Applied the same softening to
l2.induction:titlebecomes a display attribute (falls back tosignatureLabel, else"Untitled policy") whiletriggerandprocedureremain load-bearing and still fail the draft when missing.Verified with vitest: the two focused test files run 16 tests (all passing) and the full
tests/unit/memory/l2 + l3suite runs 89 tests (all passing).tsc -p tsconfig.json --noEmitis clean. Changes touch four files underapps/memos-local-plugin/{core,tests}/memory/{l2,l3}; opsp task file and openspec change artefacts are staged in.ai-tasks/andopenspec/changes/for scheduler archival.Base branch:
mainBranch:
bugfix/autodev-2335-20260903031432962Commit:
c2cf01bPR: #2338
Assigned to: @whipser030
Reviewers: @whipser030, @hijzy
github-actions commentedon Oct 4, 2026
This issue has been automatically marked as stale due to inactivity.
github-actions commentedon Oct 14, 2026
This issue has been automatically closed due to inactivity.