Skip to content

Add agent context files for alliance clusters - #111

Open
Andesha wants to merge 5 commits into
ComputeCanada:mainfrom
Andesha:alliance-agents-guidance
Open

Add agent context files for alliance clusters#111
Andesha wants to merge 5 commits into
ComputeCanada:mainfrom
Andesha:alliance-agents-guidance

Conversation

@Andesha

@Andesha Andesha commented Aug 11, 2026

Copy link
Copy Markdown

Per request, adds an AGENTS.md to the software stack to better direct agents towards best practices and documentation.

Implementation details:

  • Claude reads /etc/claude-code/CLAUDE.md. This works for users who have installed Claude on the system itself.
  • Respects symlinks. Most other harnesses and agents expect a AGENTS.md.
  • We should experiment with env vars that claude and codex (gpt/openai) respect to disable dangerous behaviour.

I am open to refining the file and testing things out on Nibi as necessary.

Comment thread agents/AGENTS.md
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
Comment thread agents/AGENTS.md Outdated
@Andesha

Andesha commented Aug 12, 2026

Copy link
Copy Markdown
Author

I've resolved the comments to what I think works. One thing that keeps coming up is adding specific links to documents. My problem with this is it's just bloat. I think we can just tell the agents to check the documentation or perhaps just say the page name. I think more opinions would be good here cause I may be wrong.

Another point is that things like the module commands, slurm commands, and these sorts of things are already known by the frontier LLMs. No need to give specific guidance on them. The low tier LLM services wont even read these docs so 🤷

should we take this to an RSNT meeting?

Comment thread agents/AGENTS.md
@Andesha
Andesha force-pushed the alliance-agents-guidance branch from 2f4fbdc to ca49c8f Compare August 13, 2026 15:01
@rahimkhoja

Copy link
Copy Markdown

my 2 CAD cents ( which are worth about 1.4 USD )

Codex: /etc/codex/skills/ <-- we should make system wide alliance skills here for codex. Since Agents is in the user dir they can remove it, but system wide skills are a bit harder
Codex: /etc/codex/requirements.toml <-- Codex main Config file

Claude: /etc/claude-code/CLAUDE.md <-- always read by all claude code's
Claude: /etc/claude-code/skills/* <-- unofficial, you put system wide skills in here, then sym link em into users dirs. this way we can centrally manage em.
Claude: /etc/claude-code/managed-settings.json Central config.. I like this , we are now installing claude system wide, and then forcing users to not be able to install it via the config file. this way we dont get hundreds of the same program installed, and we get some control

The other agents often use the same skills, so I feel we should make alliance skills and make the system have them off the bat.

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.

4 participants