What problem does it solve? When a design or specification hinges on an unverified assumption about how a runtime, library, API, or external service actually behaves, teams risk locking in broken approaches. This Skill runs a one-shot empirical probe before the decision locks and records a definitive VIABLE, NOT-VIABLE, or INCONCLUSIVE verdict. ## Core Features & Use Cases - 8-Family Probe Taxonomy: Classifies each probe into one of eight families (local script, test harness, DB query, external API capture, container exec, browser/CDP, WS/SSE, Cloudflare worker) with a matching cost and safety class. - Mandatory Safety Gates: Enforces cost-class gates (cheap-local, needs-container, needs-live-provider, needs-browser, needs-cf) so billed or shared-resource probes never run without explicit opt-in; unmet gates yield an INCONCLUSIVE verdict instead of silent skipping. - Validated Verdict Artifacts: Writes a structured VERDICT markdown file with required frontmatter and sections, validated by the bundled validate-feasibility-verdict.mjs script. - Use Case: A spec assumes a gateway forwards a query parameter. Before locking the design, run a container-exec probe, capture the actual response, and record a NOT-VIABLE verdict that forbids depending on that parameter. ## Quick Start Ask the agent to run a feasibility probe on your hypothesis, for example: verify whether the gateway forwards the provider sort parameter and write the verdict artifact to the current task folder.
