Lc From Code

/lc-from-code

End-to-end migration: scan existing code, draft or add to astra.yaml, parameterize decisions in the code, and run until everything materializes. This works both as a fresh start from code and as an augmenting pass inside an existing ASTRA project. The user's existing logic stays intact — changes should be minimal.

Invocation contexts

This skill has two invocation contexts. The first is the user-driven default described in the phases below: do the full scan → spec → parameterize → run flow.

The second is scan-only, used when /lc-from-paper's ORIENT Stage 4 invokes this skill against a cloned reference repo at work/reference/code/. The invocation prompt will tell you explicitly to do only Phase 1's scan, write the inventory to a path it specifies (typically work/reference/code-index.md), and stop — do not touch astra.yaml at the project root, do not parameterize any code, do not run anything, do not modify the cloned repo. Reach for an Explore sub-agent (or parallel Explore spawns when the repo is large enough that one survey misses the breadth) — that's the cost-effective tool for inventorying a real codebase, and there's no longer any nested-context concern that would forbid it. Trust the invocation prompt's instructions over the fresh-migration defaults below; if the prompt says scan-only, the scan-only contract holds (stop after writing the inventory file).

Phase 1: Scan & Spec

First, invoke /astra and read its Decisions section, then decide which mode applies:

• Fresh migration: no meaningful astra.yaml exists yet. Use the code scan to draft astra.yaml and universes/baseline.yaml.
• Augment existing ASTRA: astra.yaml already exists from a paper, user interview, or prior ASTRA work. Use the code scan to add to the current spec — recipes, dependencies, containers, code-backed decision options, baseline selections, implementation notes, and missing inputs / outputs where they naturally belong. Do not create a second astra.yaml, do not replace the existing structure wholesale, and surface major structure conflicts to the user before reshaping the spec.

Scanning the project

In both modes, spawn an Explore sub-agent to scan the project. Include the decision criteria in the prompt so the sub-agent can classify candidates:

Agent(subagent_type="Explore", prompt="""
Scan this project thoroughly and return a structured inventory.

For every script and notebook, report:
- File path
- What it does (read the code, don't guess)
- What files it reads (data, configs, other scripts' outputs)
- What files it writes (results, plots, models, etc.)
- Hardcoded analytical choices: magic numbers, commented alternatives,
  method-selecting branches, config dicts. Include file, line number,
  current value, and what it controls.
- How it's currently invoked (argparse, config file, nothing)

Also report:
- Dependencies (requirements.txt, pyproject.toml, environment.yml, etc.)
- Data files present in the project
- Any existing container setup (Dockerfile, Containerfile)

Return the results as a markdown table:
| Script | Purpose | Reads | Writes | Hardcoded choices |

And a separate list of ALL candidate decisions with file:line references.
Err on the side of completeness — include anything that could plausibly
be an analytical choice. The caller will filter down later.

For reference, here are the decision criteria for classifying candidates:

{paste Decisions section from `/astra` here}

""")

When the codebase is large enough that one Explore pass risks missing depth (a multi-project monorepo, a workflow folder plus a notebooks tree plus a src/ package), spawn Explores in parallel against the named subtrees — one Explore per coherent region. Aggregate their inventories into the final scan output.

Write the scan results to CLAUDE.md under ## Project Notes (fresh migration) or to the path the invocation prompt specifies (scan-only — typically work/reference/code-index.md) as a script inventory, then in fresh migration mode draft or add to astra.yaml from the scan results following the spec structure documented in /astra. In scan-only mode, stop after the inventory file lands; do not touch astra.yaml. Use the decision criteria from /astra (Decisions section) to filter candidate decisions down to only true analytical choices — most hardcoded values are implementation details, not decisions. Use current hardcoded values as defaults.

In augment mode, preserve the existing paper-derived or user-derived inputs, outputs, decisions, findings, and description unless the code scan shows a real conflict. Attach code evidence to the nearest existing home first. Create new ASTRA structure only when the code reveals a real analysis object that has no suitable home in the current spec.

For each output, list the upstream artifacts it depends on under Output.inputs: [...] and the decisions it consumes under Output.decisions: [...]. Then add a recipe.command template that references each via {inputs.} / {decisions.} and writes to {output}. Example:

outputs:
  - id: galaxy_catalog
    type: data
    inputs: [survey_data]
    decisions: [magnitude_cut, redshift_min]
    recipe:
      command: >
        python src/build_catalog.py
        --survey {inputs.survey_data}
        --magnitude_cut {decisions.magnitude_cut}
        --redshift_min {decisions.redshift_min}
        --output {output}

Also generate or update universes/baseline.yaml with all defaults matching the current hardcoded values (so the first run reproduces existing behavior).

Write to astra.yaml and universes/baseline.yaml, then validate: astra validate astra.yaml. Fix any errors.

Use AskUserQuestion to ask the user to review the spec — they can open astra.yaml directly or right-click it and open in lightcone-ui. Wait for confirmation before proceeding to implementation.

Phase 2: Implement

Parameterize the code from ASTRA decisions so the baseline run reproduces the existing behavior. The goal is minimal changes to user code. Use your best judgement for the approach — the options below are not exhaustive:

For scripts with hardcoded values: Add argparse (or extend existing argument parsing) and replace hardcoded values with the parsed args. This is the simplest case.

For notebooks: Move the .ipynb to notebooks/ (preserving it as reference), then create a .py script that does the parameterized version. The recipe points to the new script.

For config-file-driven projects: Create a thin wrapper script that accepts ASTRA decision args, writes/updates the config file, then calls the original entry point. The user's config-driven code stays untouched.

Dependencies: Check that requirements.txt includes all packages the code imports. If one doesn't exist, create it. If it's incomplete, add missing deps.

Containers: Set container: in astra.yaml based on what the scan found.

• Existing Containerfile or Dockerfile: point at it (e.g. container: Containerfile).
• No container setup but a requirements.txt: write a minimal Containerfile (FROM python:3.12-slim, copy and pip install -r requirements.txt, then COPY . .) and point container: at it.
• Nothing to go on: set container: python:3.12-slim as a starting point — the user can swap to a real Containerfile later.

Whatever approach you use:

• Don't refactor, restructure, or improve the code. Just add the parameter plumbing.
• The recipe template is what wires decisions to scripts. Each {decisions.} placeholder in recipe.command substitutes the active option ID at runtime, and the recipe author chooses the script-side flag name. Underscore IDs in the spec → match-them-yourself flags in the script (e.g. --outlier_sigma {decisions.outlier_sigma} paired with parser.add_argument('--outlier_sigma')). There is no auto-injection.
• Output paths. The recipe receives the output directory as {output} — pass that through to the script (--output {output}) and have the script write its artifact inside that directory (e.g. {output}/data.parquet).
• Update recipes in astra.yaml if the entry point or command changed.

Phase 3: Run & Debug

lc run --universe baseline

If it fails, read the error, fix it, and retry. Iterate until lc status shows all outputs as ok.

If the scan found existing results elsewhere in the project, compare them against the new outputs in results/baseline// to verify the migration preserved behavior.

Then validate the spec and the provenance chain: astra validate astra.yaml and lc verify. Present summary to user.

Rules

• Minimal changes. Do not refactor, rename, reorganize, or "improve" existing code.
• Don't guess. Read every script before making claims about what it does.
• Filter decisions aggressively. Most hardcoded values are implementation details, not analytical choices.
• Preserve behavior. The baseline universe with default values must reproduce the original behavior exactly.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: LightconeResearch
• Source: LightconeResearch/lightcone-cli
• License: BSD-3-Clause
• Homepage: https://docs.lightconeresearch.org/

Install and usage instructions live in the source repository linked above.