A persistent memory layer for Stellar and Soroban projects. Scans a repo, links its contracts to what is actually deployed on-chain, and serves the result to you and to your AI agents over MCP.
Agent Claude Code
Use for the agent-facing MCP surface — adding, renaming or reshaping MCP tools, their descriptions and input schemas, and the end-to-end MCP tests. Owns src/commands/mcp.ts and tests/mcp.test.ts. Use it whenever changing what an AI agent can ask the memory.
Agent Claude Code
Use for the vault format — Markdown note rendering, frontmatter, wikilinks, note paths, staleness handling, and the index.json schema. Owns src/store/.ts and src/core/types.ts. Use it when changing what a note looks like or how the graph is persisted.
Agent Claude Code
Use to verify the whole tool still works before a commit, a demo, or a submission. Runs typecheck, the full test suite, a clean rebuild, a real scan of the demo project, and the Soroban contract build, then reports regressions with evidence. Use it as a gate, not for making changes.
Agent Claude Code
Use for any change to how stellar-memory reads Soroban source — detecting contracts, functions, storage durability and TTL, events, errors, cross-contract calls, or auth. Owns src/scanner/rust.ts and its tests. Use it when adding a new Soroban pattern to detect, or when the analyzer reports something wrong.
Agent Claude Code
Planner and delegator for work that spans more than one part of stellar-memory — a change touching the analyser and the vault and the MCP surface, a release preparation, a multi-file refactor. Decomposes the goal, assigns each piece to the specialist that owns those files, defines the verification gate up front, and…
Agent Claude Code
Implements changes anywhere in stellar-memory that no narrower specialist owns — CLI commands, the live window and watchers, query and signals, git integration, scan orchestration, wiring and refactors. Writes the test first, matches the surrounding code, and proves the change ran. Use the owning specialist instead…
Agent Claude Code
Use for anything touching the stellar CLI bridge or on-chain data — contract interfaces, Wasm hashes, metadata, aliases, networks, drift detection. Owns src/stellar/cli.ts and src/stellar/spec.ts. Use it before assuming any stellar CLI flag or output shape.
Agent Claude Code
Read-only investigator. Use to find out how something in this repo actually works, where a behaviour comes from, whether a claim in the README or a doc is backed by code, or what a change would touch — before anyone edits anything. Breadth-first across many files; returns conclusions with file:line evidence, never…
Agent
--- name: stellar-archivist description: Use after a decision is made, a bug is understood, an approach is rejected, or a session ends with something learned. Writes that reasoning into the project's memory notes so it survives. Use it when you catch yourself thinking "we should write this down".
Agent
--- name: stellar-builder description: Use to write or change Soroban contract code in this project — adding an entry point, changing storage, wiring a cross-contract call, fixing a diagnosed bug. The only agent that edits contracts. Have the explorer or debugger establish context first.
Agent
--- name: stellar-debugger description: Use when something is wrong — a failing test, a failed contract invocation, an unexpected on-chain result, or a warning from the project memory. Works backward from the symptom through the contract graph to the cause. Use it before the builder agent, never in…
Agent
--- name: stellar-explorer description: Use to understand how this Stellar project fits together — what the contracts are, how they call each other, what a subsystem does, where something lives. Read-only. Use it before changing anything, and whenever a question starts with "how does" or "where is".
Agent
--- name: stellar-resume description: Use when returning to this project after time away, or when starting a session and needing to know where things stand. Produces a situation report — what this project is, what moved, what is deployed, what is pending, what is risky right now. Use it first, befo…