memorize

name	memorize
description	Use the memorize skill to store observations, entities, tasks, and claims into the filesystem-backed active memory. Covers session lifecycle, working-set writes, entity/task card creation, claim extraction, and candidate promotion.

Memorize

Store new knowledge into the agent's active memory layer.

When to use

• Starting or ending a working session
• Recording observations, notes, or decisions
• Registering entities or tasks encountered during work
• Extracting structured claims from unstructured text
• Promoting candidate memories to trusted status

Quick start

import pathlib
from src.api.memory_manager import MemoryManager

mgr = MemoryManager(root_dir=pathlib.Path(".agent-memory"))

# 1. Start a session
session = mgr.start_session(current_goal="investigate auth failures")

# 2. Record a working item
item = mgr.add_working_item(
    item_type="observation",
    content="The auth service returned 401 after cert rotation.",
)

# 3. Register an entity
entity = mgr.add_entity(
    name="auth-service",
    entity_type="service",
    description="Handles authentication and token issuance.",
)

# 4. Register a task
task = mgr.add_task(title="Check certificate expiry dates", priority=1)

# 5. Extract claims from text
claims = mgr.extract_claims(
    text="The auth service failed after a certificate rotation.",
    source_ref="incident-report-42",
)

# 6. Promote eligible memories
results = mgr.promote_memories()

# 7. Close the session
mgr.end_session()

Key APIs

Class	Method	Purpose
MemoryManager	start_session()	Open a new working session with an optional goal
MemoryManager	end_session()	Close the active session
MemoryManager	add_working_item()	Append an observation, note, or decision to the working set
MemoryManager	add_entity()	Write an entity card to active memory
MemoryManager	add_task()	Write a task card to active memory
MemoryManager	extract_claims()	Extract and persist structured claims from text
MemoryManager	promote_memories()	Run governance rules and promote eligible candidates
SessionManager	create_session() / close_session()	Low-level session lifecycle

Grounding

• src/api/memory_manager.py — high-level orchestration API
• src/active_memory/session_manager.py — session create / close / load
• src/active_memory/working_set.py — JSONL working-set appender
• src/active_memory/entity_card.py — entity card writer
• src/active_memory/task_card.py — task card writer
• src/claims/extractor.py — claim extraction from text
• src/governance/promotion.py — promotion eligibility engine
• tests/unit/test_active_memory.py — unit coverage
• notebooks/01_active_memory_basics.ipynb — executable walkthrough

Rules

1. Always use MemoryManager from src/api/memory_manager.py as the entry point; do not bypass it with low-level appenders unless extending the API itself.
2. Preserve the filesystem-first layout: active/, journal/, claims/ directories under the memory root.
3. Keep claim extraction aligned with the existing ClaimExtractor pipeline rather than duplicating extraction logic.
4. Validate changes with tests/unit/test_active_memory.py.