workshop-build-an-agent
AGENTS.md156 starsChanged 6 days agoApache-2.0View on GitHub
Instructions for coding agents (Codex, Claude Code, …) working in this repository. This is the Build-an-Agent workshop (a.k.a. DevX / DevX-Lab): a seven-module, hands-on course that runs on NVIDIA AI Workbench. Learners work through Jupyter notebooks under code/ and lessons under .devx/. This repo ships its tutor as Agent Skills — the cross-harness SKILL.md format. They auto-load when an agent runs from the project root and trigger by description, or can be invoked explicitly. In Codex, invoke a skill with…
- Reads credentials
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,700 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository28 other files it gives its agents.
- module-1Skill
- module-2Skill
- module-3Skill
- module-4Skill
- module-5Skill
- module-6Skill
- module-7Skill
- nvwb-projectSkill
- nvwbSkill
- setup-workshop-nemoclaw-operatorSkill
- setup-workshop-nemoclawSkill
- setup-workshopSkill
- workshopSkill
- module-1Skill
- module-2Skill
- module-3Skill
- module-4Skill
- module-5Skill
- module-6Skill
- module-7Skill
- nvwb-projectSkill
- nvwbSkill
- setup-workshop-nemoclaw-operatorSkill
- setup-workshop-nemoclawSkill
- setup-workshopSkill
- workshopSkill
- code-reviewSkill
- technical-writingSkill
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 28
When acting as the tutor (i.e.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Instructions for coding agents (Codex, Claude Code, …) working in this repository.
This is the Build-an-Agent workshop (a.k.a. DevX / DevX-Lab): a seven-module,
hands-on course that runs on NVIDIA AI Workbench. Learners work through Jupyter
notebooks under code/ and lessons under .devx/.
The workshop tutor (Agent Skills)
This repo ships its tutor as Agent Skills — the cross-harness SKILL.md format.
They auto-load when an agent runs from the project root and trigger by description, or can
be invoked explicitly. In Codex, invoke a skill with $name (e.g. $workshop); in
Claude Code use /name (e.g. /workshop):
$workshop— overview + router across the 7 modules$module-1…$module-7— one tutor per module$setup-workshop— local install helper$nvwb,$nvwb-project— NVIDIA AI Workbench CLI + in-container project awareness
(In Codex you can also run /skills to pick a skill from a menu. The /name slash form is
Claude Code only — typing /module-1 in Codex returns "unrecognized command".)
The skills exist in two parallel trees so either harness works identically:
.claude/skills/— read by Claude Code.agents/skills/— read by Codex (auto-discovered from<repo>/.agents/skills/)
When acting as the tutor (i.e. a learner invokes a $module-N or $workshop skill — /module-N in Claude Code),
follow the skill's own rules: guide, don't solve — give graduated hints, never
complete a learner's exercise, and never reveal or open answer-key files
(*answer*, answer_key/, *.answers/). The canonical tutoring policy lives in
.agents/skills/workshop/references/tutor-policy.md. These rules apply to tutoring
sessions only — they do not restrict normal development work on the repo itself.
Contributor convention: skills are a synced pair
.claude/skills/ is the canonical source. .agents/skills/ is generated from
it by .agents/sync-codex-skills.sh, which applies a small set of harness-specific
substitutions (invocation phrases like "run claude" → "run codex", and the path
.claude/skills/ → .agents/skills/) and preserves a few hand-authored Codex files
(see the HAND_AUTHORED list in that script and .agents/skills/VENDORED.md).
Edit skills in .claude/skills/, then run bash .agents/sync-codex-skills.sh to
regenerate the Codex tree. Run bash .agents/sync-codex-skills.sh --check to verify the
two trees are in sync (CI-friendly: non-zero exit on drift). Do not hand-edit generated
files in .agents/skills/ — your changes will be overwritten on the next sync.
Environment notes
- The workshop targets NVIDIA AI Workbench containers (Linux, x86_64 and aarch64).
postBuild.bashpreinstalls bothclaudeandcodex(non-fatal if npm hiccups).~/.claude/and~/.codex/are persisted across container rebuilds (see.project/spec.yaml).- API keys load from
secrets.env(NVIDIA, optional Tavily/LangSmith); the harness itself needs its own credential (codex loginorOPENAI_API_KEY; an Anthropic key forclaude).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.