I'm authorized to test [platform name] in my own environment. Audit it for security weaknesses within the scope described in [scope document]. For each suspected vulnerability, document the evidence, the trust boundary it crosses and…

Audit [project repository] at commit [commit hash] using a read-only checkout. Create one task per concrete trust-boundary question, listing the relevant files and an evidence folder for each. Run…

Help build [approval service] that sits between [agent] and [account services]. Validate every action the agent proposes against a strict schema, then apply configurable approval rules: allow safe reads automatically and require a human decision…

Audit my [service name] REST and MCP codebase in [repository] for security problems. Focus on CI triggers that untrusted contributors could use, token scopes, handling of external messages, automatic channel joins…

Review [repository] for secrets or credentials that end up in URLs, logs or client code. Find any ingestion token passed in a WebSocket query string and move authentication into a header or subprotocol handshake that…

Connect our deterministic security engine at [engine location] to a local MCP server that exposes only three kinds of tools: posture inspection, admission simulation and evidence retrieval. Reconcile the payload formats between the engine and…

Review the diff for [pull request or commit range] for security risks such as injection, auth bypass, secret exposure, unsafe deserialization and missing input validation. Skip any issue already discussed in that pull request's comments.…

Investigate [repository] for backend bugs. Set up its local build and run its existing tests, then exercise [important behaviours] in isolated environments, such as sandboxed copies of the service. Look…