Releases·Tool Release
Developer releases toolwall, a fail-closed security gateway for AI agent tool calls
A developer built toolwall, a fail-closed checkpoint that sits between an LLM's tool call and its execution, blocking dangerous calls that pass JSON schema validation. Registration acts as an allowlist, and unknown tools, schema or policy violations, budget overruns, and detected secrets are blocked before execution. The developer is inviting the community to stress-test it.
2 engagement2 sourcesFri, Aug 28, 2026, 02:06 PM
AI Overview
- toolwall enforces that only registered tools can run, making registration itself the allowlist
- Blocks unknown tools, schema violations, policy violations, budget hits, and detected secrets before execution
- Motivating example: delete_records(filter={}) is valid JSON but wipes a table; a recipient field injected from a web page can exfiltrate data
- Argues structured outputs and JSON schema validation guarantee well-formedness, not permission
- Developer is soliciting attempts to break the gateway
Entities
Related
Products·launched
Security gate that validates AI agent tool calls beyond schema checks
A policy-checking layer that sits between an LLM agent and its tool executions, catching calls that are schema-valid but wrong (overbroad deletes, external emails, leaked secrets, retry loops). The builder invites others to red-team it against their own agents. It targets developers building tool-calling agents.
A policy-checking layer that sits between an LLM agent and its tool executions, catching calls that are schema-valid but wrong (overbroad deletes, external emails, leaked secrets, retry loops). The builder invites others to red-team it against their own agents. It targets developers building tool-calling agents.
2 engagement1 sourceredditSat, Aug 29, 2026, 01:45 PM
Products·building
Pryxor — runtime security layer for AI agent tool calls
A small service that sits between an AI agent and the tools it can invoke. Agents send tool calls to Pryxor, which authenticates the agent, validates arguments against the tool's JSON schema, and evaluates the call against a deterministic policy before execution. This lets developers give agents tool access without embedding credentials in the agent itself.
A small service that sits between an AI agent and the tools it can invoke. Agents send tool calls to Pryxor, which authenticates the agent, validates arguments against the tool's JSON schema, and evaluates the call against a deterministic policy before execution. This lets developers give agents tool access without embedding credentials in the agent itself.
7 engagement1 sourceredditThu, Oct 1, 2026, 01:25 PM
Products·building
Local runtime gate that intercepts and vets AI agent tool calls before execution
An open-source pilot that sits between an LLM agent and its tools, intercepting each tool call at runtime. It checks intent, simple data provenance, and policy, then returns ALLOW or BLOCK with a reason and writes an audit log. It targets developers protecting agents from prompt-injection-driven tool calls (e.g., a hidden instruction in a document triggering unintended HTTP, email, DB, or file actions).
An open-source pilot that sits between an LLM agent and its tools, intercepting each tool call at runtime. It checks intent, simple data provenance, and policy, then returns ALLOW or BLOCK with a reason and writes an audit log. It targets developers protecting agents from prompt-injection-driven tool calls (e.g., a hidden instruction in a document triggering unintended HTTP, email, DB, or file actions).
10 engagement1 sourceredditWed, Sep 23, 2026, 01:00 PM
Products·launched
CSL-Core: open-source tool that inventories AI agents and enforces guardrails on their tool calls
CSL-Core is an open-source, locally-run tool that scans a machine for AI agents (coding assistants, MCP servers, LangChain/OpenAI Agents/CrewAI tools, cron jobs), maps what resources they can reach, and enforces policy limits on their tool calls outside the model. It solves the problem that prompt-based safeguards ('never refund more than $500') fail sometimes and leave no record. It targets developers and teams running agents with real-world tools like refunds, shell commands, file writes, and Slack posting.
CSL-Core is an open-source, locally-run tool that scans a machine for AI agents (coding assistants, MCP servers, LangChain/OpenAI Agents/CrewAI tools, cron jobs), maps what resources they can reach, and enforces policy limits on their tool calls outside the model. It solves the problem that prompt-based safeguards ('never refund more than $500') fail sometimes and leave no record. It targets developers and teams running agents with real-world tools like refunds, shell commands, file writes, and Slack posting.
7 engagement1 sourceredditFri, Oct 9, 2026, 01:09 PM
Products·launched
OpShield – open-source safety layer for AI agents: tool-call interception, rollback, cost control
OpShield is an open-source Python platform that adds a safety layer around AI agents by intercepting every tool call. It classifies risk, blocks critical actions, enforces budgets, and enables rollback via state snapshots. It targets developers building agents with OpenAI, Anthropic, LangChain, CrewAI, or AutoGen who need guardrails and audit trails.
OpShield is an open-source Python platform that adds a safety layer around AI agents by intercepting every tool call. It classifies risk, blocks critical actions, enforces budgets, and enables rollback via state snapshots. It targets developers building agents with OpenAI, Anthropic, LangChain, CrewAI, or AutoGen who need guardrails and audit trails.
3 engagement1 sourceredditMon, Oct 12, 2026, 04:52 PM