Dual-Brain Memory Guardian

Description

This skill combines a strict Markdown rule brain with a Pinecone experience brain. It is designed for reliable behavior constraints, correction-driven learning, and semantic recall of prior pitfalls.

Usage

Use this skill when any of the following applies:

1. The user corrects your output or asks for a rewrite.
2. You complete non-trivial work and need structured self-reflection.
3. You need to recall similar historical pitfalls with fuzzy semantic matching.
4. You need high-confidence behavior constraints that must not be violated.

Common trigger hints: correction, rewrite, reflection, pitfall, memory recall, behavior constraints.

Instructions

Document Ownership (Single Source of Truth)

To avoid drift and contradictory edits, documentation ownership is strict:

1. Runtime execution flow lives in operations.md only.
2. Learning criteria and promotion logic live in learning.md only.
3. Safety and data boundaries live in boundaries.md only.
4. Setup and wiring steps live in setup.md only.

If there is any wording mismatch, follow the owner document above instead of this file.

Architecture (Minimal Contract)

This skill uses a dual-brain contract:

1. Left Brain (Markdown): explicit rules and durable preferences.
2. Right Brain (Pinecone): episodic corrections, reflections, and semantic recall.

Conflict order remains fixed:

• Project Markdown > Domain Markdown > Global Markdown > Pinecone recall.

Mandatory Trigger Entry Points

The skill requires these command-level hooks to exist:

1. memory:session-start
2. memory:auto-session-start
3. memory:on-correction
4. memory:on-task-complete
5. memory:auto-task-complete
6. memory:mark-promoted

Execution details, retries, and sequencing are defined in operations.md. Runtime guard for these triggers is enforced by scripts/memory-cli.js.

Proactive Trigger Behavior (Required)

When this skill is active, trigger commands should be called proactively:

1. On session/conversation start, call memory:auto-session-start before substantial work.
2. On final response for non-trivial work, call memory:auto-task-complete before ending.
3. Keep memory:on-correction for immediate correction capture events.

Fallback policy:

1. If auto wrappers are unavailable, call memory:session-start and memory:on-task-complete manually.

Quick Reference

| Topic | File | |------|------| | Skill contract | SKILL.md | | Setup | setup.md | | Runtime operations | operations.md | | Reflection template | reflections.md | | Heartbeat behavior | heartbeat-rules.md | | Safety boundaries | boundaries.md | | Pinecone config/runtime | src/pinecone/ | | CLI entrypoint | scripts/memory-cli.js |

Requirements

• Node.js >= 20
• npm
• @pinecone-database/pinecone
• Environment variable: PINECONE_API_KEY
• Pinecone integrated index model: multilingual-e5-large

Optional for bulk import:

• Object storage path (s3://, gs://, Azure Blob URL)
• Integration ID for private buckets

Rules Index

• Learning and promotion: learning.md
• Runtime triggers and recall flow: operations.md
• Local reflection template: reflections.md
• Safety and redaction: boundaries.md
• Scale and compaction strategy: scaling.md

Scope

This skill ONLY:

• Maintains rule memory in ~/dual-brain-memory-guardian/.
• Maintains experience memory in Pinecone.
• Uses npm-based Pinecone SDK operations (upsertRecords, searchRecords, startImport, describeImport, describeIndexStats).

This skill NEVER:

• Treats Pinecone recall as stronger than explicit Markdown contracts.
• Stores sensitive raw secrets in vector memory.
• Performs destructive heartbeat rewrites of uncertain content.

Feedback

• If useful: clawhub star dual-brain-memory-guardian
• Keep skills updated: clawhub sync