AI-assistant context files for the REMIND
codebase, loaded by Claude Code through a CLAUDE.md pointer (see Install).
AGENT.md— single source of truth: repo-wide instructions (model overview, HPC environment, debugging protocol, renv workflow, …). Imports the topic files below via@-references.instructions/— topic-specific deep dives (gams.md,r.md,cicd.md,gdx.md).prompts/— on-demand task prompts (triage-failed-run.md,inspect-verify-gdx.md). Not auto-loaded; invoked explicitly as slash commands when needed.scripts/— reusable tools referenced by the prompts (e.g.gdx_inspect.R, a context-safe GDX inspector).
remind-agent lives inside each REMIND checkout as a plain untracked
clone. A one-line CLAUDE.md in the REMIND root points to it; Claude Code
reads CLAUDE.md (not AGENT.md) and resolves @-imports transitively:
<remind>/
├── CLAUDE.md <- one line: @remind-context/AGENT.md
├── remind-context/ <- this repo
│ ├── AGENT.md
│ ├── instructions/
│ ├── prompts/
│ └── scripts/
└── ...
Both stay out of REMIND's git status via .git/info/exclude. A pointer, not a
copy, so pulling this repo updates the context.
In the root of each REMIND checkout:
git clone git@github.com:remindmodel/remind-agent.git remind-agent
echo '@remind-agent/AGENT.md' > CLAUDE.md
printf 'remind-agent/\nCLAUDE.md\n' >> .git/info/excludeIf the checkout already has a CLAUDE.md, add the @remind-agent/AGENT.md
line to it instead of overwriting it.
Pull the latest context into a checkout:
git -C remind-agent pullcd remind-agent
git checkout main
# edit AGENT.md / instructions/*.md / prompts/*.md
git commit -am "..."
git pushPrompts in prompts/ are on-demand, not auto-loaded.
To invoke one as /<name>, add a shim .claude/commands/<name>.md in REMIND
containing @../../remind-agent/prompts/<name>.md, or just open the prompt
file and paste it.