code-create-staged-plan

Turn a substantial change into a dependency-ordered staged plan with a gate at each step. Plans live in notes/<codebase>/; scoping evidence lives in dated agent_logs/.

Read an existing notes/vllm/ plan before writing a new one.

When to use

• A feature port (upstream → our fork), a multi-step fix (esp. with a parity/regression requirement), a refactor, a kernel/perf change, or any change too big or too risky to land in one commit.
• NOT for a one-line fix or a mechanical edit — just do those (and log if non-obvious).

Where it lives

• Plan and scope docs: /Users/benjaminfeuer/Documents/notes/<codebase>/ — one parent (README.md or <change>_plan.md) and one stage<N>_<slug>_scope.md per stage.
• Evidence: dated /Users/benjaminfeuer/Documents/agent_logs/YYYY-MM-DD_<topic>.md; cite it from the plan.

Parent plan doc — required sections

1. Header: date · status (scoped — propose-only; no code yet at creation) · target repo + canonical local path + isolated working-copy path (the rsync-clone you'll branch in — see "Isolated working copy" below) + branch (the feature branch you'll cut in the clone, e.g. feuer/<slug>) · links to the evidence agent_logs/.
2. Goal: the precise, testable end state (e.g. "dcp=N rollout bit-identical to dcp=1: greedy token-ids identical + logprobs allclose atol 1e-2").
3. Mechanism: root cause or design rationale.
4. Stage map: Stage | title | what | feature(s) | layer | cost (CPU / 1-GPU / N-GPU) | gate. Each stage must be independently testable and build on a gated predecessor; mark the critical path.
5. Global invariants (assert in EVERY stage): the flag-off / default-off byte-identical contract (a new feature is a no-op until its flag flips — mirror the EP/CP scaffold no-op tests); the parity gate (the load-bearing equivalence, e.g. G2 bit-identical); regression bounds (don't break MLA / the other arms); minimal diff (no gratuitous API/config churn).
6. Borrow map (don't reinvent): the exact files/functions/line-anchors you'll touch or copy from — and a standing note that anchors DRIFT (reconfirm at impl time; they're from a dated read).
7. Safety / reward-hacking (where relevant): policy-invariance for RL reward shaping, ground-truth anchors, "down-weight not zero", parse-real-signals-only.
8. Validation discipline: per-stage, what proves the gate (flag-off byte-identical first → behavior-on test → GPU smoke on the correct SIF/env). Name the measurement (paired McNemar + pass@k, torch.equal, allclose tol at the bf16 floor — don't loosen a tol silently).

Per-stage scope doc (stage<N>_<slug>_scope.md) — required sections

• Header: date · status (scoped GO / blocked / …) · companion = the parent · "no fix yet" if scope-only.
• Why this is the next step.
• Change-set: exactly what files change (or "test-only; no <repo>/ source touched this stage").
• Validation gate (GO/NO-GO): the concrete pass condition + cost. This is what code-execute-staged-plan checks before advancing.
• Composes with / depends on: the upstream stages it assumes are already green.

Isolated working copy — rsync-clone BEFORE you branch (do NOT branch the canonical clone)

Use this rsync-clone path only for penfever/working self-merge repos: ~/Documents/{OpenThoughts-Agent,vllm}. Marin forks (harbor, MarinSkyRL, evalchemy) use git-worktree→PR→main. Never cut a feature branch on the canonical clone; rsync it to an isolated working directory first:

# 1. rsync-clone the canonical repo (INCLUDING .git; skip heavy build/venv dirs) to an isolated copy
SLUG=<change-slug>; REPO=OpenThoughts-Agent          # or vllm  (marin-forks harbor/MarinSkyRL/evalchemy use worktree→PR→main instead)
SRC=/Users/benjaminfeuer/Documents/$REPO
DST=/Users/benjaminfeuer/Documents/staged-work/$SLUG/$REPO
mkdir -p "$(dirname "$DST")"
rsync -a --exclude='.venv' --exclude='__pycache__' --exclude='*.egg-info' --exclude='wandb/' "$SRC/" "$DST/"
# 2. confirm the canonical clone is clean + on penfever/working, then branch IN THE CLONE
cd "$DST" && git fetch origin && git checkout penfever/working && git pull --ff-only && git checkout -b feuer/$SLUG

• Keep the canonical clone on penfever/working.
• Record staged-work/<slug>/<repo> in the plan header; execute, commit, and push from it, then git pull other clones.
• The rsync clone is not the editable-installed copy; test clusters after push+pull. Use mcp__ide__getDiagnostics on clone files.
• Use one clone per touched repo; remove staged-work/<slug>/ after merge.

Discipline

• Local clone is ground truth: branch in an rsync clone; execution commits, pushes, and syncs rather than patching clusters.
• Default-off: flag-off is byte-identical.
• Cheapest repro first: stage 0 is usually a unit/CPU harness.
• At creation, plans are propose-only; hand off to code-execute-staged-plan for execution.