What problem does it solve? When a task needs an external tool, connector, or MCP, it is easy to assume a documented tool is actually available, run unbounded actions, or silently change shared state. This Skill enforces a disciplined workflow that verifies real runtime exposure, scopes risk, obtains authorization, and records outcomes honestly. ## Core Features & Use Cases - Capability-first resolution: Define the required capability before picking a vendor, then compare only actually exposed tools by scope, risk, permissions, and fallback. - Bounded probe execution: Plan and run read-only, sandboxed, reversible probes with explicit targets, approvals, rollback paths, and verification steps. - Learning and record boundary: Distinguish usage notes, candidate lessons, reviewed lessons, and tool-use recipes so one attempt is never promoted to shared guidance automatically. - Use Case: A user asks whether a deployment connector is available and wants a status check. The Skill verifies the runtime actually exposes the connector, runs a read-only probe against the authorized workspace, verifies the result, and records a usage note without mutating any catalog records. ## Quick Start Use the tool-discovery-and-safe-execution skill to check whether a GitHub MCP is available in this runtime and run a bounded read-only probe against my repository.
