| name | rumil-load-run |
| description | Load a rumil run into the conversation by run ID. Shows a tree of every call in the run (id, type, status, cost, event summary) grouped under the run. Use when you have a trace URL like /traces/<run_id> and want to see the whole run, not a single call. For one call's verbatim trace use rumil-trace instead. Pass --compare <other_run_id> to A/B two runs side-by-side. |
| allowed-tools | Bash |
| argument-hint | <run_id> [--full] [--only llm_exchange] [--last-n N] [--compare <other_run_id>] |
rumil-load-run
Loads every call that shares a run_id — the anchor the frontend uses for
/traces/<run_id>. Resolves short (8-char) or full run IDs.
Default output is a compact tree: one line per call with short id, type,
status, cost, duration, and a one-line event summary. Use this first to
find the interesting call, then drill into it with rumil-trace <short_id>.
Flags:
--full — also print each call's full trace (events + verbatim exchanges)--only <event> — filter events in the per-call summary--last-n N — in --full mode, trim each call to its last N exchanges--compare <other_run_id> — render the two runs side-by-side, aligned by
call_type. Within each call_type, calls are paired by creation order
(1st of-type-in-A vs 1st of-type-in-B, etc); extras land as — cells.
Rows where the two sides differ are flagged with * and a brief reason
(status, cost delta >25%, exchange count). No verbatim exchanges in compare
mode — call summaries only (id, status, cost, event counts, fruit/conf).
Use when A/B'ing two runs of similar shape (e.g. same workflow on different
prefix variants, or two judging runs on different pairs).
Note: some runs (standalone dispatches) have no row in the runs table —
they exist only as a calls.run_id tag. This skill queries calls directly
so it works for both cases. If a runs row exists, its name + config are
printed as a header.
setopt no_glob 2>/dev/null; set -f; PYTHONPATH=.claude/lib uv run python -m rumil_skills.load_run $ARGUMENTS
