Code Create Staged Plan

What it does

When your agent uses it

• The user says scope/plan this change
• Design before coding
• A change is too big/risky for one shot

Example prompts

• “scope/plan this change”
• “stage it out”
• “design before coding”
• “/code-create-staged-plan”

Workflow steps

8 steps, taken from the first numbered list in SKILL.md.

1. 1Header: date · status (scoped — propose-only; no code yet at creation) · target repo + canonical local path + isolated working-copy path…
2. 2Goal: the precise, testable end state (e.g. "dcp=N rollout bit-identical to dcp=1: greedy token-ids identical + logprobs allclose atol…
3. 3Mechanism: root cause or design rationale.
4. 4Stage map: Stage | title | what | feature(s) | layer | cost (CPU / 1-GPU / N-GPU) | gate. Each stage must be independently testable and…
5. 5Global invariants (assert in EVERY stage): the flag-off / default-off byte-identical contract (a new feature is a no-op until its flag…
6. 6Borrow map (don't reinvent): the exact files/functions/line-anchors you'll touch or copy from — and a standing note that anchors DRIFT…
7. 7Safety / reward-hacking (where relevant): policy-invariance for RL reward shaping, ground-truth anchors, "down-weight not zero"…
8. 8Validation discipline: per-stage, what proves the gate (flag-off byte-identical first → behavior-on test → GPU smoke on the correct…

What it can do on your machine

Read from SKILL.md and the folder at commit 3bd1917. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Shell commands in SKILL.md call:

  • git
  • rsync

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  No URLs in SKILL.md. Its commands use git and rsync, which can reach the network depending on how they are called.

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Code Create Staged Plan loads about 1.5k tokens when it runs. Until then it costs about 194 tokens; SKILL.md has 560 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~194

When it runs · the whole SKILL.md, loaded when a task matches

~1.5k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from open-thoughts/OpenThoughts-Agent at commit 3bd1917, republished under its Apache-2.0 licence (© open-thoughts). 560 words, ~1,463 tokens.

name

code-create-staged-plan

description

DESIGN a non-trivial codebase change (Harbor / MarinSkyRL / vLLM / OT-Agent / LLaMA-Factory) as a dependency-ordered STAGED PLAN before writing code — a feature port, a multi-step fix with parity requirements, a refactor, a kernel/perf change. Produces a parent plan doc + per-stage scope docs under notes/<codebase>/ (each stage = scope + GO/NO-GO validation gate + cost), with global invariants (flag-off byte-identical, parity gates), a borrow-map of code anchors (which drift), and safety considerations. Evidence/scoping breadcrumbs go in dated agent_logs/. Use when the user says "scope/plan this change", "stage it out", "design before coding", or a change is too big/risky for one shot. Pairs with code-execute-staged-plan (which runs the plan).

# 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

© open-thoughts, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .agents/skills/code-create-staged-plan of open-thoughts/OpenThoughts-Agent.

Open the folder on GitHubat commit 3bd1917