Claw Self Improving Plus

Build a conservative learning pipeline. Optimize for signal, not clutter.

Core stance

Do not auto-rewrite long-term memory or behavior files by default.

Use this flow:

1. 1. Capture raw learning candidates.
2. Normalize them into a structured schema.
3. Score each item for promotion value.
4. Detect duplicates or merge candidates.
5. Consolidate repeated learnings into stronger records.
6. Build a prioritized learning backlog.
7. Draft anchored candidate patches.
8. Review patches with human approval.
9. Apply only approved patches.

Learning types

Use these types:

• - mistake: the agent did something wrong
• INLINECODE1: the user corrected a wrong assumption or behavior
• INLINECODE2: a useful fact about environment, tools, preferences, or workflow
• INLINECODE3: a durable preference, policy, or chosen design
• INLINECODE4: a known failure mode that should not recur

Minimal record schema

Store each learning candidate as JSON with these fields:

• - id: stable slug or timestamped id
• INLINECODE6
• INLINECODE7
• INLINECODE8
• INLINECODE9
• INLINECODE10
• INLINECODE11
• INLINECODE12
• INLINECODE13
• INLINECODE14
• INLINECODE15
• INLINECODE16
• INLINECODE17

Default enums:

• - confidence: INLINECODE19
• INLINECODE20: INLINECODE21
• INLINECODE22: INLINECODE23
• INLINECODE24: INLINECODE25

Routing rules

Promote by destination, not vibes:

• - SOUL.md: durable style, personality, voice rules
• INLINECODE27: operating rules, workflows, safety/process lessons
• INLINECODE28: environment-specific commands, paths, model/tool preferences
• INLINECODE29: important long-term facts about user, projects, decisions, history
• daily/raw store only: low-confidence or highly local observations

If a learning does not clearly deserve promotion, keep it in the raw log.

Scoring heuristic

Score each record on five dimensions:

1. 1. reuse_value: will this help again?
2. INLINECODE31: how well supported is it?
3. INLINECODE32: how broadly does it matter?
4. INLINECODE33: should it become a lasting rule or memory?
5. INLINECODE34: where should it go if promoted?

Use this practical rubric:

• - High promotion priority: repeated mistake, explicit user preference, environment fact that breaks tasks, regression with real cost
• Medium priority: useful workflow pattern seen more than once
• Low priority: one-off trivia, speculative interpretation, emotional noise, temporary state

Anchored patch generation

Prefer anchored insertion or exact replacement over blind append.

Each patch may contain:

• - INLINECODE35
• INLINECODE36
• INLINECODE37
• INLINECODE38
• INLINECODE39
• INLINECODE40
• INLINECODE41
• INLINECODE42

Use exact replacement when the old text is known.
Use anchored insertion when the destination section is known.
Use append only as fallback.

Learning store layout

Use a stable .learnings/ structure. See references/learning-store-layout.md.

Recommended files:

• - INLINECODE45
• INLINECODE46
• INLINECODE47
• INLINECODE48
• INLINECODE49
• INLINECODE50

Default workflow

1. Capture

Append raw learnings into .learnings/inbox.jsonl.

Use scripts/capture_learning.py to create normalized records.

2. Score

Run scripts/score_learnings.py on the inbox or a batch export.

3. Review duplicates

Run scripts/merge_candidates.py to group likely duplicates.

4. Draft patches

Run scripts/draft_patches.py to produce anchored reviewable patch candidates.

5. Review

Use scripts/review_patches.py to list, approve, reject, or skip candidates.

Examples:

CODEBLOCK0

6. Apply only after approval

Run scripts/apply_approved_patches.py.

This script only applies entries explicitly approved.
It validates allowed targets, supports --dry-run, skips duplicate entries already present, and prefers exact replacement, then anchored insertion, then append fallback.

Output style

When reporting results, use this structure:

• - new_candidates: count
• INLINECODE60: count
• INLINECODE61: count
• INLINECODE62: short bullet list
• INLINECODE63: yes

Resources

References

• - Scoring rubric: see INLINECODE64
• Patch target guide: see INLINECODE65
• Learning store layout: see INLINECODE66

Scripts

• - INLINECODE67
• INLINECODE68
• INLINECODE69
• INLINECODE70
• INLINECODE71
• INLINECODE72
• INLINECODE73
• INLINECODE74
• INLINECODE75
• INLINECODE76
• INLINECODE77
• INLINECODE78
• INLINECODE79
• INLINECODE80
• INLINECODE81