memory-feedback

# Memory Feedback Skill

Give feedback on memories to help the system learn and improve.

## Feedback Types

### Positive Reinforcement (Reinforce)

When a memory was helpful or correct:

```python
from memory_lib import update_memory

# Reinforce a memory (increases confidence)
update_memory(
    memory_id="2026-01-29-pnpm-preference",
    outcome="accepted"
)
```

Effect:
- Confidence +0.1
- positive_reinforcement count +1

### Negative Feedback (Correct)

When a memory was wrong or outdated:

```python
from memory_lib import update_memory

# Correct a memory (decreases confidence)
update_memory(
    memory_id="2026-01-29-npm-default",
    outcome="rejected"
)
```

Effect:
- Confidence -0.2
- negative_reinforcement count +1
- If confidence < 0.3: status changes to "under_review"

### Add New Evidence

Add new observations to a memory:

```python
from memory_lib import update_memory
from datetime import datetime

update_memory(
    memory_id="2026-01-29-pnpm-preference",
    new_evidence={
        "timestamp": datetime.utcnow().isoformat() + "Z",
        "description": "User used pnpm again",
        "observation_id": "obs-456"
    },
    confidence_delta=0.05  # Optional small boost
)
```

### Supersede a Memory

When a new memory replaces an old one:

```python
from memory_lib import update_memory

# Mark old memory as superseded
update_memory(
    memory_id="2026-01-28-npm-default",
    outcome="superseded"
)
```

Effect:
- Status changes to "superseded"
- The superseding memory should reference this one

## CLI Feedback Commands

```bash
# Reinforce (positive feedback)
ccmem reinforce <memory_id>

# Correct (negative feedback)
ccmem correct <memory_id>

# Via sync script
python3 ~/.claude/plugins/ccmem/scripts/sync-claude-md.py --promote
```

## Automatic Feedback from Observations

The system can detect feedback signals from observations:

```python
def detect_feedback_from_observation(observation):
    """
    Analyze observation for implicit feedback.
    ""