diff --git a/README.md b/README.md
index a8de937..ceec2f5 100644
--- a/README.md
+++ b/README.md
@@ -669,15 +669,26 @@ let test_steps := conc.ask(test_actor, Execute(test_task))
 used to hardcode one — implement `list.zip`, in four fixed phases, with its
 own sequential runner — and now drives the same agent graph the TUI does.
 
+This entry point needs the same `--max-steps` override the Quickstart's
+`bin/lex-code` wrapper supplies for the TUI, and for the same reason: it's
+trusted, long-running orchestration code, not the untrusted-sandboxed-snippet
+case the VM's 10,000,000-step default guards against. There's no wrapper
+script for this entry point, so it has to be typed by hand every time —
+reproduced live: a real multi-module build ran for 17+ minutes across dozens
+of agent turns and hit `step limit exceeded` with the flag omitted, discarding
+that whole run's output (`bootstrap/run.lex`'s own step-by-step printing only
+happens after the graph run returns, so a mid-run panic here loses everything,
+not just the final result).
+
 ```sh
 # the original demo, unchanged
-lex run --allow-effects … src/bootstrap/run.lex main
+lex run --max-steps 20000000000 --allow-effects … src/bootstrap/run.lex main
 
 # a real task, phases of your choosing
 LEX_TASK="add a retry wrapper to src/http.lex" \
 LEX_PIPELINE=build,test \
 LEX_PROVIDER=litellm \
-  lex run --allow-effects … src/bootstrap/run.lex main
+  lex run --max-steps 20000000000 --allow-effects … src/bootstrap/run.lex main
 ```
 
 | Variable | Default | Meaning |
@@ -706,7 +717,7 @@ verified_on = []                      # "<path>:<kind>" — a pass on that path
 
 ```sh
 LEX_TASK_SPEC=examples/tasks/zip.task \
-  lex run --allow-effects … src/bootstrap/run.lex main
+  lex run --max-steps 20000000000 --allow-effects … src/bootstrap/run.lex main
 ```
 
 The spec's `goal` becomes the task the agents are told, so the words they act
diff --git a/bin/lex-code-bootstrap b/bin/lex-code-bootstrap
new file mode 100755
index 0000000..3f259bb
--- /dev/null
+++ b/bin/lex-code-bootstrap
@@ -0,0 +1,29 @@
+#!/usr/bin/env sh
+# Wrapper for `src/bootstrap/run.lex main` — requires `lex` CLI on PATH.
+#
+# Same reason `bin/lex-code` exists for the TUI: the VM's default step
+# budget (10,000,000, a DoS guard for *untrusted* sandboxed snippets per
+# `lex run --help`) kills a long agentic session outright with a
+# step-limit-exceeded panic once enough turns or verbose-enough model
+# output accumulate. bootstrap/run.lex's own build/test/verify pipeline
+# is exactly this case — trusted, long-running orchestration code, not
+# an untrusted snippet — and it has no wrapper of its own, so the flag
+# has to be remembered by hand every time `lex run` is typed directly.
+# Reproduced live, repeatedly: real multi-module builds running 10-20+
+# minutes across dozens of agent turns hit `step limit exceeded` with
+# the flag omitted, discarding that whole run's output (bootstrap/run.lex
+# only prints per-step output after the graph run returns, so a mid-run
+# panic here loses everything, not just the final result — see the
+# README's "Bootstrap Script" section for the full story).
+#
+# bootstrap/run.lex's `main` takes no positional args (LEX_TASK,
+# LEX_TASK_SPEC, LEX_PIPELINE, LEX_PROVIDER are all read from the
+# environment, not argv — see its own header comment for why), so unlike
+# bin/lex-code there is no `main --` separator or "$@" forwarding to get
+# right here; just the effect grant and the step-limit override.
+set -e
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+LEX_CODE_EFFECTS="${LEX_CODE_EFFECTS:-approval,concurrent,crypto,env,fs_read,fs_walk,fs_write,io,llm,net,proc,random,sql,stream,time}"
+LEX_CODE_MAX_STEPS="${LEX_CODE_MAX_STEPS:-20000000000}"
+exec lex run --max-steps "$LEX_CODE_MAX_STEPS" --allow-effects "$LEX_CODE_EFFECTS" \
+  "$SCRIPT_DIR/../src/bootstrap/run.lex" main