---

name: enumeration description: "Configure and validate REINVENT4 enumeration workflows, peptide amino-acid libraries, seed files, and attachment-point chemistry." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Enumeration

Use this sub-skill when a task mentions enumeration.toml, run_type = "enumeration", peptide enumeration, amino-acid libraries, scaffolds.smi, warheads.smi, attachment points, LibInvent, LinkInvent, Mol2Mol seed files, or PepInvent masked peptide files.

Read First

• references/enumeration-workflow.md for the run_type = "enumeration" config shape, peptide mask enumeration, scoring reuse, CLI invocation, and output CSV checks.
• references/seed-files-and-chemistry.md for scaffold, warhead, Mol2Mol, PepInvent, and enumeration amino-acid library file conventions.
• references/troubleshooting.md for attachment-point, separator, peptide-mask, config-key, scoring-block, and output failures.
• scripts/validate_seed_files.py for safe no-run validation of enumeration configs and seed files before launching REINVENT4.

Fast Path

1. Confirm the job is run_type = "enumeration"; use the sibling sampling sub-skill for generator sampling and the sibling scoring sub-skill for standalone score-only jobs.
2. Put smiles_file, amino_acid_library_file, aa_names_column, smiles_column, optional batch_size, and optional output_csv under [parameters].
3. Keep peptide templates in smiles_file as one CHUCKLES-like row per line with ? masks and | fragment separators. Runtime peptide enumeration supports one or two masks per template row.
4. Add a normal [scoring] block to score enumerated peptides; validate the scoring block with the scoring sub-skill if components or transforms are complex.
5. Run a static check before execution:

   python sub-skills/enumeration/scripts/validate_seed_files.py enumeration.toml --kind enumeration
   

6. Run a small CPU job first:

   reinvent --device cpu --seed 123 --log-filename enumeration.log enumeration.toml
   

Scope Boundaries

• This sub-skill owns run_type = "enumeration", peptide/amino-acid enumeration parameters, seed-file conventions, attachment-point sanity checks, output CSV expectations, and chemistry/library-design validation.
• For scoring component design, transforms, aggregation, and plugin discovery, use the sibling scoring sub-skill.
• For run_type = "sampling" with LibInvent, LinkInvent, Mol2Mol, or PepInvent models, use the sibling sampling sub-skill after validating seed-file shape here when attachment points or masks are the problem.
• For transfer learning, staged learning, model vocabulary, invalid generated tokens, or trained-agent checkpoints, use the sibling learning or sampling sub-skill.