fix: filter ghost observations with no content fields (#1625)

When the LLM is overwhelmed by large context it can emit bare
<observation/> blocks (or ones containing only <type>). These are
stored as rows where title, narrative, facts and concepts are all
null/empty, appearing as meaningless "Untitled" entries in the context
window. Add a guard in parseObservations() that skips any observation
where every content field is null/empty before pushing it to the
result array.

Generated by Claude Code
Vibe coded by ousamabenyounes

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
Ousama Ben Younes
2026-04-10 09:40:40 +00:00
parent 3651a34e96
commit e398212983
2 changed files with 170 additions and 3 deletions
+15 -3
View File
@@ -50,9 +50,8 @@ export function parseObservations(text: string, correlationId?: string): ParsedO
const files_read = extractArrayElements(obsContent, 'files_read', 'file');
const files_modified = extractArrayElements(obsContent, 'files_modified', 'file');
// NOTE FROM THEDOTMACK: ALWAYS save observations - never skip. 10/24/2025
// All fields except type are nullable in schema
// If type is missing or invalid, use first type from mode as fallback
// All fields except type are nullable in schema.
// If type is missing or invalid, use first type from mode as fallback.
// Determine final type using active mode's valid types
const mode = ModeManager.getInstance().getActiveMode();
@@ -83,6 +82,19 @@ export function parseObservations(text: string, correlationId?: string): ParsedO
});
}
// Skip ghost observations — records where every content field is null/empty.
// These accumulate when the LLM emits a bare <observation/> (or one with only <type>)
// due to context overflow. They carry no information and pollute the context window.
// (subtitle and file lists are intentionally excluded from this guard: an observation
// with only a subtitle is still too thin to be useful on its own.)
if (!title && !narrative && facts.length === 0 && cleanedConcepts.length === 0) {
logger.warn('PARSER', 'Skipping empty observation (all content fields null)', {
correlationId,
type: finalType
});
continue;
}
observations.push({
type: finalType,
title,