Read recorded agent runs instead of guessing history

This skill is the judgement layer over OrcaReplay's MCP server. OrcaReplay itself records a coding-agent run below the harness and can replay it offline or fork it onto another model; this skill's job is narrower - it tells an agent when to stop guessing about a past session and go read the recording instead. Its core rule: when a question is about something that already happened, read the trace before answering, rather than reconstructing it from memory or from a transcript you were handed. Both of those are missing the tool results, the exit codes, and the files that changed without anyone mentioning it - a recording is evidence, memory and transcripts are not, and guessing is the wrong move even when the guess would have been right.

Use it whenever a question concerns what actually happened in a prior recorded agent run, and a recording exists under .orca/runs to check instead of reconstructing events from memory. It depends entirely on OrcaReplay already being set up: the orcareplay npm package (Node 20+) installed, its MCP server registered as orca, and at least one recording present under .orca/runs - without those, there is nothing for this skill to route to.