By tslateman
Explicit context management for multi-agent systems — capture decisions, patterns, and failures across projects
Maps structural relationships within and across projects. Invoke when: scanning a codebase for undocumented workflows, auditing the knowledge graph for orphans and missing edges, tracing data flows across system boundaries, checking contract coverage, or discovering entry points and state machines. Unlike lore-scribe (records from conversation) and lore-keeper (retrieves on demand), this agent proactively surveys codebases and audits the graph for structural gaps.
Deep context retrieval from Lore. Invoke when: asking "what do we know about X?", needing background on a topic or decision, understanding why something was built a certain way, finding related decisions or patterns, exploring the knowledge graph. Unlike lore resume (ambient context), this agent performs targeted deep retrieval following graph edges and reading related entries.
Resolves entity identity across Lore's knowledge graph post-hoc. Invoke when: finding duplicate nodes, auditing graph coherence, detecting contradictions between decisions, wiring orphan nodes, or cleaning up superseded entries. Unlike lib/conflict.sh (blocks duplicates at write time), this agent audits what already exists -- scanning the full corpus for semantic overlaps, contradictions, and structural gaps.
Records decisions and patterns to Lore. Invoke after: architectural decisions (e.g., "let's use PostgreSQL"), tool or library selections, rejection of alternatives (e.g., "we tried X but it didn't work because Y"), discovering reusable patterns, making non-obvious implementation choices. A decision was just made when the conversation contains phrases like "let's go with", "we'll use", "the right approach is", "I chose X over Y", or when an alternative was explicitly rejected with rationale.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Explicit context management for multi-agent systems.
# Clone the repository
git clone https://github.com/tslater/lore.git ~/dev/lore
# Add to PATH (add to ~/.bashrc or ~/.zshrc)
export PATH="$HOME/dev/lore:$PATH"
# Verify installation
lore --help
Lore stores user data (decisions, patterns, sessions) separately from tool code. Run the install script to set this up:
./scripts/install.sh
This creates ~/.local/share/lore/ and migrates any existing data from the
repo. Add to your shell profile:
export LORE_DATA_DIR=~/.local/share/lore
brew install jq yq
# Debian/Ubuntu
sudo apt install jq sqlite3
pip install yq
# Or via snap
sudo snap install yq
# One verb, four destinations — flags determine type
lore capture "Users retry after timeout" # → observation (inbox)
lore capture "Use JSONL for storage" --rationale "Append-only, simple" # → decision (journal)
lore capture "Safe bash arithmetic" --solution 'Use x=$((x+1))' # → pattern (patterns)
lore capture "Permission denied" --error-type ToolError # → failure (failures)
# Shortcuts still work
lore remember "Use JSONL for storage" --rationale "Append-only, simple"
lore learn "Safe bash arithmetic" --solution 'Use x=$((x+1))'
lore fail ToolError "Permission denied"
# End a session with handoff notes
lore handoff "Auth implementation 80% complete, need OAuth integration"
# Resume previous session at start of new session
lore resume
# Search across all components
lore search "authentication"
# Graph-enhanced search (follows relationships)
lore search "authentication" --graph-depth 2
Run lore help for all commands, or lore help <topic> for details on capture, search, intent, registry, or components. See the tutorial for a hands-on walkthrough and docs/engram-integration.md for the Engram bridge.
MEMORY.md gives agents implicit context—loaded into the prompt, hoped to be relevant. Lore provides explicit context—structured writes, typed queries, cross-project assembly. This matters when:
Two retrieval phases, used together:
--graph-depth 1-3)lore resume at session start loads the previous session's state—what was done, what's next, what's blocked. lore handoff at session end captures context for the next agent.
Journal captures decisions with rationale. Patterns capture lessons learned. Graph connects concepts. These components earn their keep at scale—when flat files stop fitting in a prompt.
| Component | Purpose | Key Question |
|---|---|---|
| journal/ | Decision capture | "Why did we choose this?" |
| patterns/ | Lessons learned | "What did we learn?" |
| transfer/ | Session succession | "What's next?" |
| graph/ | Knowledge connections | "What relates to this?" |
| registry/ | Project metadata | "What exists and how does it connect?" |
| intent/ | Goals and specs | "What are we trying to achieve?" |
| inbox/ | Raw observations | "What did we notice?" |
| evidence/ | Factual evidence | "What supports this?" |
User data lives at $LORE_DATA_DIR (default: ~/.local/share/lore):
~/.local/share/lore/
├── journal/data/ # decisions.jsonl
├── patterns/data/ # patterns.yaml
├── transfer/data/ # sessions/*.json
├── graph/data/ # graph.json
├── intent/data/ # goals/
├── inbox/data/ # observations.jsonl
├── failures/data/ # failures.jsonl
├── evidence/data/ # evidence.jsonl
└── search.db # FTS5 index, embeddings, graph cache
Run lore init to scaffold this structure, or scripts/install.sh to
migrate existing data.
Projects integrate via lib/lore-client-base.sh—fail-silent wrappers that record decisions and patterns without blocking if Lore is unavailable. See LORE_CONTRACT.md for the full interface.
npx claudepluginhub tslateman/claude-plugins --plugin loreClaude Code plugin for reflection, code quality, writing, and design
tslatemancmux sidebar status, tool progress, and notifications for Claude Code sessions
Two-tier agent orchestration: Manager decomposes and delegates, Workers execute focused subtasks
Protocol for project memory and execution discipline. Structures how knowledge, decisions, research, specs, work, evidence, and reviews are captured across sessions and agents.
z3z1maPure filesystem-based memory system for Claude Code with custom ontology support
zircoteOpenLTM — Long-Term Memory for AI coding agents: semantic search, context injection, session learning
RohiRIKPersistent memory for AI coding agents -- captures tool usage, compresses via LLM, injects context into future sessions. 12 hooks, 41 MCP tools, 4 skills, real-time viewer.
rohitg00Persistent agent memory that survives across sessions — auto-compacting 3-tier memory with hybrid search. Your agent remembers what it learned, decided, and built.
kevin-hs-sohnPersistent memory for AI coding agents. Survives across sessions and compactions.
Gentleman-Programming