Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/nvidia-nemo/gym/gh-stacknpx skills add NVIDIA-NeMo/Gym --skill gh-stackgit clone --depth 1 https://github.com/NVIDIA-NeMo/GymWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00065 | $0.09057 |
| Opus 5 | $0.00032 | $0.04529 |
| Sonnet 5 | $0.00013 | $0.01811 |
| Haiku 4.5 | $0.00006 | $0.00906 |
Grade A, and why
gh-stack scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 857 lines — stays where its author put it; the contents beside it link to each section on GitHub.
gh-stack
gh stack is a GitHub CLI extension for managing stacked branches and pull requests. A stack is an ordered list of branches where each branch builds on the one below it, rooted on a trunk branch (typically the repo's default branch). Each branch maps to one PR whose base is the branch below it, so reviewers see only the diff for that layer.
main (trunk)
└── feat/auth-layer → PR #1 (base: main) - bottom (closest to trunk)
└── feat/api-endpoints → PR #2 (base: feat/auth-layer)
└── feat/frontend → PR #3 (base: feat/api-endpoints) - top (furthest from trunk)
The bottom of the stack is the branch closest to the trunk, and the top is the branch furthest from the trunk. Each branch inherits from the one below it. Navigation commands (up, down, top, bottom) follow this model: up moves away from trunk, down moves toward it.
When to use this skill
Use this skill when the user wants to:
- Break a large change into a chain of small, reviewable PRs
- Create, rebase, push, or sync a stack of dependent branches
- Navigate between layers of a branch stack
- View the status of stacked PRs
- Tear down and rebuild a stack to remove, reorder, or rename branches
Prerequisites
The GitHub CLI (gh) v2.0+ must be installed and authenticated. Install the extension with:
gh extension install github/gh-stack
Before using gh stack, configure git to prevent interactive prompts:
git config rerere.enabled true # remember conflict resolutions (skips prompt on init)
git config remote.pushDefault origin # if multiple remotes exist (skips remote picker)
Agent rules
All gh stack commands must be run non-interactively. Every command invocation must include the flags and positional arguments needed to avoid prompts, TUIs, and interactive menus. If a command would prompt for input, it will hang indefinitely.
- Always supply branch names as positional arguments to
init,add, andcheckout. Running these commands without arguments triggers interactive prompts. - When a prefix is set, pass only the suffix to
add.gh stack add authwith prefixfeat→feat/auth. Passingfeat/authcreatesfeat/feat/auth. - Always use
--autowithgh stack submitto auto-generate PR titles. Without--auto,submitprompts for a title for each new PR. - Always use
--jsonwithgh stack view. Without--json, the command launches an interactive TUI that cannot be operated by agents. There is no other appropriate flag — always pass--json. - Use
--remote <name>when multiple remotes are configured, or pre-configuregit config remote.pushDefault origin. Without this,push,submit,sync,link, andcheckouttrigger an interactive remote picker. - Avoid branches shared across multiple stacks. If a branch belongs to multiple stacks, commands exit with code 6. Check out a non-shared branch first.
- Plan your stack layers by dependency order before writing code. Foundational changes (models, APIs, shared utilities) go in lower branches; dependent changes (UI, consumers) go in higher branches. Think through the dependency chain before running
gh stack init. - Use standard
git addandgit commitfor staging and committing. This gives you full control over which changes go into each branch. The-Amshortcut is available but should not be the default approach—stacked PRs are most effective when each branch contains a deliberate, logical set of changes. - Navigate down the stack when you need to change a lower layer. If you're working on a frontend branch and realize you need API changes, don't hack around it at the current layer. Navigate to the appropriate branch (
gh stack down,gh stack checkout, orgh stack bottom), make and commit the changes there, rungh stack rebase --upstack, then navigate back up to continue. - Use
gh stack linkfor external tool workflows. When branches are managed by an external tool (jj, Sapling, etc.), usegh stack link branch-a branch-b.linkdoes not rely on local tracking state and is intended for API-driven PR and stack management. Always provide at least 2 branch names or PR numbers.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 857 lines · 65 tokens per session scan A 088d8e9895ad
gh-stack is a skill published in the GitHub repository NVIDIA-NeMo/Gym (1,148 stars, last pushed 3d ago), licensed Apache-2.0. It adds 65 tokens to every session and 9,057 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
writing-bench-task-judge
Use when writing or modifying checkgoals() / getanswer() / App check methods in benchenv/task/, or when reviewing a draft task's judge correctness. Triggers include adding a new task, editing a judge method, or diagnosing a judge false-positive/negative.
testing-bench-task
Use when adding or modifying offline judge tests for benchenv tasks — specifically entries in OFFLINEJUDGEPOSITIVECASES / OFFLINEJUDGENEGATIVECASES in benchenv/tests/ /testtasks.py, or writing live tests. Triggers after a new task is added, or when tightening judge coverage.
designing-bench-task
Use when designing a new benchenv task suite, adding several new tasks to an existing suite, or critiquing a task-set proposal for a mobile-gym App — before any class FooTask(...) is written under benchenv/task/.
project-graveyard
Scans the developer's machine for dead side projects, autopsies each one from its git history (died at the payments wall, killed by a newer project, finished but never shipped), surfaces their personal death patterns, and picks the corpse most worth resurrecting — then helps ship it. Use when the user mentions…
web-app-penetration-testing
Pentest a web app or website end to end — black-box testing of a live URL, staging environment, or local dev server that finds and exploits real vulnerabilities (auth bypass, broken access control, IDOR, injection, XSS, SSRF, business logic) and proves each one with a working proof-of-concept instead of a signature…
haiku
When writing a haiku for this bot, follow these conventions.