Path-rooted sub-analysis recipes execute from project root, breaking sub-analysis-relative paths

[lhparker1]

What happened

Recipes in a path-rooted sub-analysis execute from the project root, but lc init scaffolds a per-sub-analysis scripts/ dir and the engine renders {output} as root-relative (e.g. analyses/reconstruction/results/<universe>/<output_id>). A recipe authored the natural way — python scripts/run_reconstruction.py, with the script reading cwd-relative paths like ../../data/... — fails because there is no scripts/ at the project root and the relative data path does not resolve. The scaffold/docs treat a sub-analysis as a self-contained unit (cwd = the sub-analysis dir), but the executor runs every recipe from the project root with no cd into the analysis_path.

Error

python: can't open file '<project_root>/scripts/run_reconstruction.py': [Errno 2] No such file or directory

Reproduction

1. Create a path-rooted sub-analysis (analyses/<name>/ with its own astra.yaml and scripts/).
2. Author a recipe: command: python -u scripts/foo.py --output {output}.
3. lc run <name>.<output_id> from the project root.
4. Resolved command runs from project root: the script path and any cwd-relative reads (../../data/...) fail.

The engine renders {output} root-relative (analyses/<name>/results/...) yet runs from the project root — internally inconsistent with sub-analysis-relative authoring.

Suggested resolution: either cd into the path-rooted sub-analysis dir and render {output} relative to it, or document/scaffold that ALL recipe paths must be project-root-relative.

Environment

• ASTRA (astra-spec): 0.0.7
• lightcone-cli: 0.3.0
• Python: 3.11.14
• OS: Linux 6.4.0 (NERSC Perlmutter)

[lhparker1]

In my own words -- if you try to materialize an output of a sub-analysis from the root directory, the sub-analysis astra.yaml might have the python file defined relative to its local subdirectory, instead of relative to the root, which would then break relative paths etc. So I guess a fix is to either encourage everything to be defined relative to the parent root, or to cd in whenever things are run from a sub-analysis. I guess this might also be solved if the sub-analysis is stored in the upstream astra yaml.