orca-replay

1# Reading a recorded agent run
2 
3A recording is evidence. Your memory of a session is not, and neither is a transcript you were
4handed — both are missing the tool results, the exit codes, and the files that changed without
5anyone mentioning it.
6 
7**The rule: when a question is about something that already happened, read the trace before you
8answer.** Do not reconstruct it. If a recording exists, guessing is the wrong move even when the
9guess would have been right.
10 
11## When to Use This Skill
12 
13- "Why did you delete/overwrite/move X?"
14- "What changed this file?" / "Which step broke the build?"
15- "Can you reproduce yesterday's failure?"
16- "Does this still reproduce?" (see the limit on that in step 4 — replay cannot tell you
17  whether a *fresh* run would fail again)
18- "Would a different model have got this right?"
19 
20## Workflow
21 
22### 1. Find the run
23 
24`orca_list_runs` — newest first, and it names the run each fork came from. Skip this only when the
25user clearly means the most recent one; every other tool defaults to `run: "last"`.
26 
27### 2. Narrow to the chain that produced the thing being asked about
28 
29`orca_show_run` gives the whole timeline: model turns with token counts and stop reasons, tool
30calls with arguments and results, shell commands with exit codes, and every file the run changed.
31Good for orientation, long for a specific question.
32 
33`orca_graph` is usually the better tool. It returns causal edges — which event produced which. Pass
34…