agency-memory

/agency-memory

Query and manage the Great Minds agency vector store. Memories are stored in SQLite with embeddings for semantic search.

Usage

/agency-memory search "query text"
/agency-memory add --type learning "content here"
/agency-memory add --type decision --agent "Steve Jobs" --project "Dash" "content"
/agency-memory list --type board-review
/agency-memory list --agent "Jensen Huang"
/agency-memory import
/agency-memory export
/agency-memory stats

Instructions

When this skill is invoked, run the memory CLI from the memory-store directory.

Search

Find relevant memories by semantic similarity:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory search "$QUERY" --limit ${LIMIT:-5}

Add

Store a new memory. Valid types: learning, decision, qa-finding, board-review, retrospective, architecture.

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory add \
  --type "$TYPE" \
  --agent "$AGENT" \
  --project "$PROJECT" \
  --content "$CONTENT"

List

List memories filtered by type, agent, or project:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory list --type "$TYPE"
cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory list --agent "$AGENT"
cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory list --project "$PROJECT"

Import

Import existing markdown memories from the Great Minds project:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory import

This parses:

• MEMORY.md index (key learnings)
• memory/*.md files (operational learnings, architecture decisions, research)
• rounds/*/decisions.md (locked project decisions)
• rounds/*/board-review-*.md (Jensen Huang board reviews)
• rounds/*/qa-report-*.md (Margaret Hamilton QA findings)

Export

Export all memories back to markdown for compatibility:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory export

Stats

Show memory store statistics:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory stats

Maintenance

Prune

Remove duplicate memories (cosine similarity > 0.92) and low-value entries (e.g., "all green" board reviews, "PASS" QA findings):

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory prune
cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory prune --threshold 0.85

Consolidate

Find clusters of similar memories (similarity > 0.75) and merge them into single stronger entries:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory consolidate
cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory consolidate --threshold 0.80

Optimize

Score all memories by recency, uniqueness, and richness. Remove the bottom 10% by score:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory optimize
cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory optimize --percentile 20

Maintain

Run the full maintenance cycle (prune + consolidate + optimize) in sequence. This is what the dream cron calls automatically:

cd "${PIPELINE_REPO:-$(git rev-parse --show-toplevel)}/memory-store" && ./bin/memory maintain

Environment

• OPENAI_API_KEY — enables neural embeddings (text-embedding-3-small). Falls back to TF-IDF when absent.
• MEMORY_DB — custom path to SQLite database. Defaults to memory-store/memory.db.

Bundled Store

The memory store also ships with this plugin at memory-store/. Any installation gets a fresh copy. Install dependencies before first use:

cd "/path/to/great-minds-plugin/memory-store" && npm install

Design

• SQLite with WAL mode for concurrent reads
• Cosine similarity search over embedding vectors
• Content-hash embedding cache to avoid redundant API calls
• Works offline via TF-IDF fallback
• Runs locally and on DigitalOcean with no external services