ruflo-agentdb

Reviewed at commit b9f3245

Caution: Its skills pre-approve Bash and have Claude run the unpinned npm package @claude-flow/cli@latest through npx, so each new version runs without asking.

Substrate plugin for Ruflo memory: AgentDB controller bridge (15 agentdb_* MCP tools), RuVector ONNX embeddings (10 embeddings_* tools incl. RaBitQ 32x…

Made by
ruvnet
What we checked

Refuses writes to the Ruflo memory tools when they contain what looks like a key, token or password, and can check files passed to a memory import. If you switch recall on (off by default), it looks up your prompt in the connected Ruflo memory server and adds the best matches to the prompt as data. It runs no programs and makes no network calls of its own.

Review details

Static review of the source at the pinned commit; not run. Delta review of the changes since 00432b4; the rest carries over from the review at that commit. Changes in 0.4.13 touch no hook code: the version bump, a repository field and two keywords in plugin.json, README and CHANGELOG text, a pinned-version mention (ruvector 0.3.12) in agents/agentdb-specialist.md, version strings in scripts/smoke.sh (dev script), and a line in skills/agentdb-query/SKILL.md describing the agentdb_graph-pathfinder modes (personalized-pagerank, dynamic-mincut). No new tools, allowed-tools, commands or hooks. Carried over: registers session.start (finds the project root, registers /agentdb-mod, writes .claude-flow/agentdb-mod/status.json via $.fs.write), prompt.submit (only when the recall option is on and source is not none), tool.call (only when guard is on, the default) and command.run for agentdb-mod. Recall lists connected tools with $.tool.list and calls memory_search, agentdb_hierarchical-recall, agentdb_pattern-search or hooks_recall on the already connected MCP server through $.mcp.call, using the first 1000 characters of the prompt, a deadline and a 10 minute in-memory cache; results are screened for secrets and injection phrasing, capped at 5 items and 1500 characters, and attached as a retrieved-memory block marked as data. The guard denies calls to known memory writer tools whose input matches secret patterns; for memory_import it reads the named file (at most 1 MB, only under the project root or $HOME) and refuses when it holds a secret; it fails open on any error. The status file holds counters and up to five 120-character snippets. The two skills (agentdb-query, vector-search) list plain Bash in allowed-tools, and the commands, agent and skills suggest unpinned npx @claude-flow/cli@latest commands (memory init --force, hooks post-task, embeddings and memory CLI). Not reviewed: the ruflo-core MCP server, the npm package, scripts/smoke.sh beyond the diff.

What it uses
  • Reads your prompts: yes
  • Sees tool calls: yes
  • Blocks or changes tool calls: yes
  • Reads or writes files: yes
  • Runs programs on your machine: no
  • Calls the model (uses tokens): no
  • Sends data over the network: no
  • Stores data: yes
  • Shows something on screen: yes
Hooked events

Substrate plugin for Ruflo memory: AgentDB controller bridge (15 agentdb_* MCP tools), RuVector ONNX embeddings (10 embeddings_* tools incl. RaBitQ 32x quantization), and WASM HNSW pattern router (3 ruvllm_hnsw_* tools). As a mod (ADR-445): opt-in safe recall into the prompt, a write guard that keeps secrets out of memory, /agentdb, and a status file the console shows

Made this mod? Add a badge to your README

modsforClaude badge for ruflo-agentdb

Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.

[![Listed on modsforClaude](https://modsforclaude.com/badge/ruflo-agentdb.svg)](https://modsforclaude.com/mods/ruflo-agentdb/)

← All mods