[wan9chi]

Motivation

Since #966, vp run --dry-run --json fails for a task whose command runs vp run …. The dry-run RFC shows such a step as expandedFrom with its own plan, so the run plan is a tree, and each nested plan has its own dependencies.

Changes

// "nested": "vp run build"
{ "expandedFrom": "vp run build",
  "plan": { "build": { "package": ".", "name": "build", "steps": [ { "program": "…", "cache": { "status": "hit", … } } ] } } }

• The layout recurses into nested graphs. Each plan's dependsOn is scoped to its own tasks, so the same task ID can appear at several levels.
• expandedFrom is the item's command as written, with any args given after the task name.
• A leaf's step is found by its path through the nested plans.
• The error for tasks that run vp run goes away.

A nested vp run --dry-run is still a spawned step, which a dry run doesn't start. The e2e fixture run_plan_json_nested covers plans at several levels, args forwarded to a nested run, and a nested dry run.

🤖 Generated with Claude Code

---

Note

Medium Risk
Changes the --dry-run --json output shape and planning/reporter behavior for nested runs; consumers of that JSON and anyone relying on the previous hard error should validate against the new nested step format.

Overview
vp run --dry-run --json no longer errors when a task invokes nested vp run … commands. Those steps are emitted as nested plan nodes (expandedFrom plus an embedded plan) instead of aborting.

The JSON reporter builds a tree of plans: it walks expanded execution graphs, scopes each plan's dependsOn to that plan only (so the same task ID can appear at multiple depths), and maps leaf cache results into command steps via a path through nested plans. The run-plan schema and Rust document types split steps into command vs nested variants; the changelog reflects the new behavior.

E2E coverage adds multi-level nested plans, argument forwarding into nested runs, and the case where a nested vp run --dry-run stays a plain spawned step (not inlined).

^{Reviewed by Cursor Bugbot for commit 1d46e32. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

fspy benchmark

linux

dynamic/launch             change  -0.10%  [ -3.42% ..  +3.69%]  overhead   +55.56%
dynamic/access             change  +0.00%  [ -2.89% ..  +3.94%]  overhead   +20.50%
dynamic/access-relative    change  +0.00%  [ -2.43% ..  +3.06%]  overhead   +66.68%
dynamic/access-contended   change  +1.03%  [ -5.42% ..  +5.86%]  overhead   +49.91%
static/launch              change  +0.31%  [ -2.13% ..  +3.14%]  overhead  +199.14%
static/access              change  -0.17%  [ -1.35% ..  +0.76%]  overhead +1449.79%
static/access-relative     change  +0.30%  [ -0.99% ..  +1.43%]  overhead +1922.38%
static/access-contended    change  -0.29%  [ -1.16% ..  +0.87%]  overhead +2091.75%

macos

dynamic/launch             change  +0.60%  [ -5.60% ..  +5.62%]  overhead  +260.77%
dynamic/access             change  +0.35%  [ -6.96% .. +10.30%]  overhead    +5.58%
dynamic/access-relative    change  +1.87%  [ -2.96% .. +11.14%]  overhead  +263.38%
dynamic/access-contended   change  -3.11%  [-86.88% ..  +8.80%]  overhead    +5.95%

windows

dynamic/launch             change  +0.22%  [ -1.36% ..  +1.62%]  overhead   +29.73%
dynamic/access             change  +0.00%  [ -0.63% ..  +0.64%]  overhead    +1.09%
dynamic/access-relative    change  +0.00%  [ -0.61% ..  +0.61%]  overhead    +1.27%
dynamic/access-contended   change  +0.41%  [ -0.66% ..  +1.19%]  overhead    +1.66%

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 480cb60. Configure here.}