Auto-approve from mem7

agent-mesh queries mem7 for past approval decisions before submitting to the approval queue. If a tool+agent pattern has enough consistent approvals (default: 3+, 0 rejections), the request is auto-approved.

How it works

Tool call → policy: human_approval
  │
  ├─ Level 0: Policy engine (static rules, instant)
  │
  ├─ Level 1: Built-in mem7 lookup (~100ms)
  │   query mem7 for past decisions (tool + agent + tags=["decision"])
  │   3+ approvals, 0 rejections → auto-approve (supervisor:mem7)
  │   else → escalate
  │
  ├─ Level 1+: External supervisor (Python, rules + Ollama, ~20s)
  │   polls approval queue, evaluates, resolves
  │   else → escalate
  │
  └─ Level 2: Human (terminal prompt or agent7 UI)

Each auto-approved decision is written back to mem7, reinforcing the pattern for future queries.

Configuration

memory:
  url: http://localhost:9070    # mem7 daemon URL
  token: ""                     # optional Bearer token

supervisor:
  auto_approve: true            # default true when memory.url is set
  min_approvals: 3              # threshold (default 3)

Set auto_approve: false to disable even when mem7 is configured.

Example: testing end-to-end

Prerequisites: agent-mesh v0.9.1+, mem7 running on :9070.

1. Verify mem7 is reachable

curl -s http://localhost:9070/rpc \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | head -c 100

2. Seed 3 approval decisions

Simulate 3 past approvals for filesystem.write_file by agent claude:

for i in 1 2 3; do
  curl -s -X POST http://localhost:9070/rpc \
    -H "Content-Type: application/json" \
    -d "{
      \"jsonrpc\": \"2.0\",
      \"id\": $i,
      \"method\": \"tools/call\",
      \"params\": {
        \"name\": \"memory_store\",
        \"arguments\": {
          \"key\": \"decision.filesystem.write_file.test${i}\",
          \"value\": \"approved by user:marc — agent:claude tool:filesystem.write_file reason:routine write\",
          \"tags\": [\"decision\", \"approved\", \"filesystem.write_file\", \"agent:claude\"],
          \"agent\": \"agent-mesh\"
        }
      }
    }"
done

3. Call the tool via HTTP

curl -s -X POST http://localhost:9090/tool/filesystem.write_file \
  -H "Authorization: Bearer agent:claude" \
  -H "Content-Type: application/json" \
  -d '{"params":{"path":"/home/user/test.txt","content":"auto-approve test"}}' \
  | python3 -m json.tool

Expected: no approval prompt, immediate response with "policy": "allow".

4. Check the trace

curl -s "http://localhost:9090/traces?tool=filesystem.write_file&limit=1" \
  | python3 -m json.tool

Expected trace entry:

{
    "agent_id": "claude",
    "tool": "filesystem.write_file",
    "policy": "allow",
    "policy_rule": "supervisor:mem7",
    "latency_ms": 3
}

policy_rule: "supervisor:mem7" confirms the built-in Level 1 supervisor resolved the request. The original policy was human_approval, but mem7 had 3 prior approvals — so it passed without blocking.

5. Verify the decision was written back

curl -s -X POST http://localhost:9070/rpc \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "memory_search",
      "arguments": {
        "query": "decision filesystem.write_file",
        "tags": ["decision"],
        "limit": 5
      }
    }
  }' | python3 -m json.tool

You should see 4 entries: the 3 seeded + 1 auto-approval by supervisor:mem7.

Behavior summary

Scenario	Action	Traced as
3+ approvals, 0 rejections	Auto-approve	supervisor:mem7
Any rejections in history	Escalate	Normal approval flow
Not enough history	Escalate	Normal approval flow
mem7 down or unreachable	Escalate	Normal approval flow
auto_approve: false	Skip check	Normal approval flow

Graceful degradation

• mem7 unreachable → escalate (3s timeout, never blocks)
• mem7 returns error → escalate
• Search returns no results → escalate
• All failure modes fall back to the normal approval flow — the auto-approve is additive, never subtractive.

Metrics

Monitor the mem7 write path at GET /metrics:

agent_mesh_mem7_writes_attempted_total
agent_mesh_mem7_writes_succeeded_total
agent_mesh_mem7_writes_failed_total

A growing failed count means mem7 is down — auto-approve will escalate everything until it recovers.