MemClawz Connect

Connect any AI agent to the MemClawz shared memory bus. Gives agents read-before-act and write-after-complete patterns via a simple HTTP API. Use when an age...

Security Scan

The SKILL.md is coherent with a shared-memory connector, but it omits required environment declarations, comes from an unknown source, and instructs the agent to send arbitrary data to an arbitrary/remote MemClawz URL (including a public IP example) — this combination raises data-exfiltration and provenance concerns.

Purpose & Capability

The skill's stated purpose (connect an agent to a MemClawz HTTP memory bus) matches the instructions (curl-based read/write/search), but metadata lists no required environment variables while the runtime instructions clearly require MEMCLAWZ_URL and MEMCLAWZ_AGENT_ID (and optionally MEMCLAWZ_API_KEY). The omission of these expected requirements and the lack of a homepage/origin (source: unknown) are inconsistent and reduce trust.

Instruction Scope

SKILL.md instructs the agent to run curl against whatever MEMCLAWZ_URL is set to (localhost or any remote host) and to write arbitrary 'content' after tasks. That behavior is expected for a memory connector, but it also allows sending agent context and results to any remote endpoint the operator configures (the doc even shows a public IP example). The instructions do not restrict what data to include, do not require authentication by default, and do not warn about sensitive data — increasing the risk of accidental or malicious exfiltration.

Install Mechanism

There is no install spec and no code files (instruction-only). This minimizes local attack surface because nothing is downloaded or written by the skill itself.

Credentials

Declared requirements list no environment variables or credentials, but the runtime docs require MEMCLAWZ_URL and MEMCLAWZ_AGENT_ID and mention MEMCLAWZ_API_KEY if auth is enabled. This mismatch is problematic: required secrets/vars are not declared in metadata, and a skill that can send data to an arbitrary URL should explicitly declare and justify any credentials it needs.

Persistence & Privilege

The skill is not always-enabled and uses the default model-invocation policy (agent may invoke it autonomously). Autonomous invocation plus the ability to send arbitrary data to a remote MemClawz server increases blast radius if the server is untrusted — but autonomous invocation alone is expected and not a standalone red flag.

What to consider before installing

This skill probably does what it claims, but there are missing declarations and provenance issues you should address before installing. Specifically: (1) Verify you trust and control the MemClawz server you point MEMCLAWZ_URL at — do not set it to an unknown public IP. (2) Treat MEMCLAWZ_URL, MEMCLAWZ_AGENT_ID, and MEMCLAWZ_API_KEY as required configuration and only provide them if you understand where data will be stored. (3) Ask the publisher for a homepage, source code, and a clear security model (auth, data retention, encryption). (4) If you plan to use this with agents that may handle secrets or PII, restrict the skill to agents you control or modify the AGENTS.md procedure to sanitize or exclude sensitive content before writing. (5) Prefer a deployment where MEMCLAWZ_URL is localhost or an internal, access-controlled host, and require an API key on the server side. If the vendor cannot justify the missing metadata and unknown origin, avoid enabling it for agents handling sensitive data.

One skill. Any agent. Shared memory.

Setup

export MEMCLAWZ_URL="http://localhost:3500"   # or remote: http://YOUR_SERVER:3500
export MEMCLAWZ_AGENT_ID="my-agent"           # unique per agent

No API key required for default installs. If auth is enabled, also set MEMCLAWZ_API_KEY.

Health Check

curl -s "$MEMCLAWZ_URL/health"
# {"status":"ok","version":"...","qdrant":"connected"}

Agent Protocol

Before ANY Task — Search First

curl -s "$MEMCLAWZ_URL/api/v1/search?q=TOPIC&limit=5"

Response:

{"results": [{"content": "...", "agent_id": "quant-dev", "memory_type": "decision", "score": 0.92}]}

Use results as context before starting work. Avoids re-discovering what's already known.

After Completing Work — Write Back

curl -s -X POST "$MEMCLAWZ_URL/api/v1/add" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Deployed v2.0 — fixed auth race condition with mutex on refresh",
    "agent_id": "'"$MEMCLAWZ_AGENT_ID"'",
    "memory_type": "event"
  }'

Memory Types

Type	When
fact	Discovered info (endpoints, versions, configs)
decision	Choices made (architecture, approach, tool selection)
procedure	How something was done (deploy steps, build process)
event	What happened (deployed X, fixed Y, shipped Z)
insight	Lessons learned (what worked, what didn't)
intention	Planned actions
commitment	Promises made
action	Actions taken
outcome	Results of actions

Stats

curl -s "$MEMCLAWZ_URL/api/v1/stats"

List Agents

curl -s "$MEMCLAWZ_URL/api/v1/agents"

Get Memories

curl -s "$MEMCLAWZ_URL/api/v1/memories?agent_id=$MEMCLAWZ_AGENT_ID&limit=50"

AGENTS.md Integration

Append to your agent's AGENTS.md:

## MemClawz Shared Memory

Fleet memory API: $MEMCLAWZ_URL/api/v1

### Before ANY task:
Search shared memory for relevant context:
curl -s "$MEMCLAWZ_URL/api/v1/search?q=<task keywords>&limit=5"

### After completing ANY significant work:
Write results to shared memory:
curl -s -X POST $MEMCLAWZ_URL/api/v1/add \
  -H "Content-Type: application/json" \
  -d '{"content": "<what was done>", "agent_id": "$MEMCLAWZ_AGENT_ID", "memory_type": "<type>"}'

Remote Agents

For agents on a different server, just change MEMCLAWZ_URL from localhost:3500 to the master's IP/hostname:

export MEMCLAWZ_URL="http://76.13.154.71:3500"

Everything else stays the same.