Claude Code skills that answer one question each, with evidence.
/plugin marketplace add abovebeyond-ai/plugins
What's running on its own in your repo - and has it earned that? Finds everything that acts without a human typing - agents, LLM calls, MCP servers, bots, scheduled jobs - and scores it on the PAC framework: autonomy earned or configured, blast radius, reliability with an error margin.
/plugin install agent-scan@abovebeyond
/agent-scan
What happens when your model makes something up? Two modes on the ProveML method. Hand it a document and it verifies every claim against your data: each fact traced to a source, each number sourced (an unsourced number is a finding even when it happens to be right), each judgment checked against a declared threshold. Hand it nothing and it audits the repo: every path where model output reaches a human or a decision, scored by the gate in between - nothing, review, structure, or verification.
/plugin install hallucination-scan@abovebeyond
/hallucination-scan
Every scan obeys one house rule: prove, don't claim - each level assigned comes
with file:line evidence, or is reported as unknown. MIT licensed, by
Above Beyond.