Skip to content

Plan ↔ receipt lab: the run's two bookends - #10

Draft
arielconti10 wants to merge 2 commits into
ariel/cloud-agent-1783295266127-g5fcifrom
ariel/plan-receipt-lab
Draft

Plan ↔ receipt lab: the run's two bookends#10
arielconti10 wants to merge 2 commits into
ariel/cloud-agent-1783295266127-g5fcifrom
ariel/plan-receipt-lab

Conversation

@arielconti10

Copy link
Copy Markdown
Owner

What

New playground lab at /playground/plan-receipt (stacked on #9).

An agent run has two bookends with the same content model: the plan is the prospective account ("here's what I'll touch, how, and when I'll stop"), the receipt the retrospective one. The lab's centerpiece is the pattern no shipping product does well: a receipt that diffs against the plan.

  • Plan phase — the contract as a document-like artifact (approach, scope, stopping condition, budget, exclusions), mirroring Cursor's plan mode rather than composer chrome. Dispatch when it matches intent.
  • Running phase — high-autonomy simulation; only the boundary escalation pauses ("the clean fix bumps a dep, but the ask excludes deps"), with two choices whose consequences differ downstream.
  • Receipt phase — each delivered item classified against the plan: "as planned" or "deviation — escalation feat: migrate from Vite to Next.js 16 App Router #1" with the cause line ("You allowed one devDependency at the boundary escalation"). Choose the workaround instead and the receipt reports zero deviations with the delivered fix reflecting the rewrite. Stopping condition and budget reconcile planned vs delivered with met / under-cap chips.

Reuses the shared run script from run-surface-data.ts (extended with a RUN_PLAN structure) and the same fixture ask, so the three labs — composer dressing, run surface, plan↔receipt — form one continuous story.

Why

Completes the bookend thesis from the four-surface framework: deviations are acceptable when traced to a human decision, and the receipt proves the tracing, not just the delta. A receipt without a plan can only list what happened; it can never reconcile delivery against intent. This is the candidate content model for a future run-receipt registry primitive.

Verification

Typecheck, lint, audit-ui, Prettier all pass. Browser QA at desktop and 390px: both escalation branches, run-again loop, plan phase. Playground-only — no registry changes.

Open in Web Open in Cursor 

@vercel

vercel Bot commented Jul 6, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
agentic-craft Ready Ready Preview, Comment Jul 6, 2026 10:48am

Request Review

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant