Menu

1 / 5

Every AI toolchain is inventing its own safety layer. We shipped one that works across all of them.

DEV Community·Sunil Prakash·6 months ago
#W2Khttn5
#ai#security#opensource#jamjet#audit
Reading 0:00
15s threshold

Same policy. Three runtimes.

# Claude Code, with @jamjet/claude-code-hook installed as a PreToolUse hook: > Delete the old customer records from the staging DB.

Tool request: bash.shell_exec Args: psql -c "DELETE FROM customers WHERE created_at < '2024-01-01'"

JamJet policy: BLOCKED (rule: shell.exec) Audit: ~/.jamjet/audit/2026-05-11/claude-code-hook.jsonl

Enter fullscreen mode

Exit fullscreen mode

# OpenAI Agents SDK (TS), with @jamjet/openai-guardrail wired into a refund tool: JamjetPolicyBlocked: JamJet policy: BLOCKED (tool: payments.refund, rule: payments.*)

Audit: ~/.jamjet/audit/2026-05-11/openai-guardrail.jsonl

Enter fullscreen mode

Exit fullscreen mode

# Claude Desktop talking to a Postgres MCP server, fronted by @jamjet/mcp-shim: {"jsonrpc":"2.0","id":7,"error":{ "code": -32000, "message": "JamJet policy: BLOCKED (rule: *delete*)", "data": {"tool": "postgres.delete_all_rows", "audit": "mcp-shim.jsonl"} }}

Enter fullscreen mode

Exit fullscreen mode

Same policy.yaml . Three runtimes.…

Continue reading — create a free account

Join HashtagPLUS to read full articles, follow hashtags, vote, and join the conversation.