iiRecord
Agentic AI Atlas · GAP-STATE-001: Long-Term Memory Extraction
page:docs-harness-features-backlog-gaps-state-continuity-gap-state-001a5c.ai
II.
Page reference

page:docs-harness-features-backlog-gaps-state-continuity-gap-state-001

Reading · 1 min

GAP-STATE-001: Long-Term Memory Extraction reference

Extract and persist key learnings, decisions, and patterns from completed runs into a long-term memory store that informs future orchestration sessions.

Pagewiki/docs/harness-features-backlog/gaps/state-continuity/GAP-STATE-001.mdOutgoing · 0Incoming · 1

GAP-STATE-001: Long-Term Memory Extraction

FieldValue
Categorystate-continuity
PriorityHigh
EffortL
StatusMissing

Description

Extract and persist key learnings, decisions, and patterns from completed runs into a long-term memory store that informs future orchestration sessions.

Current State

No memory extraction. State is per-run journal only. harness:retrospect analyzes but does not persist extracted memories.

Target State

Memory extraction from completed runs into ~/.a5c/memory/. Memories indexed by project, topic, and recency. Extracted memories injected into new run prompts as context.

Dependencies

Key Files

ComponentPath
Journal storagepackages/sdk/src/storage/
Session managementpackages/sdk/src/session/
Prompts modulepackages/sdk/src/prompts/

Recommendation

Phase 2 implementation. Extend harness:retrospect to extract and persist key learnings. Index by project and topic. Inject into new run prompts.