---

name: relaxation-and-outputs description: "Post-process ColabFold results with Amber/OpenMM relaxation, confidence/output validation, plotting utilities, citations, and extra pTM/interface metrics." disable-model-invocation: true metadata: disco-role: operating license: MIT

Relaxation and Outputs

Use this sub-skill when the task starts after ColabFold input preparation, MSA generation, or structure prediction and asks you to inspect, validate, relax, visualize, or interpret prediction outputs.

Route here for

• Running standalone Amber/OpenMM relaxation with colabfold_relax on existing PDB outputs.
• Choosing CPU vs GPU relaxation and tuning --max-iterations, --tolerance, --stiffness, or --max-outer-iterations.
• Explaining output files such as ranked/unrelaxed/relaxed PDBs, score JSON files, PAE JSON, pLDDT/PAE/coverage PNGs, config.json, and cite.bibtex.
• Validating an output directory without re-running prediction by using scripts/inspect_colabfold_outputs.py.
• Interpreting pLDDT, pTM, ipTM, PAE, ranking metrics, and optional extra interface metrics.
• Using plotting or display APIs such as plot_predicted_alignment_error(...), plot_msa_v2(...), and show_pdb(...) when optional visualization dependencies are available.
• Deciding citation coverage for AlphaFold, multimer, MMseqs/MSA, templates, environment database, DeepFold, and Amber relaxation.

Route elsewhere

• For FASTA, CSV, A3M, PDB/mmCIF input preparation, use ../inputs-and-formats/SKILL.md.
• For MSA server, local MMseqs2 database, GPU search, or AF3 JSON-only MSA export workflows, use ../msa-search/SKILL.md.
• For running colabfold_batch, model selection, templates during prediction, or prediction GPU/JAX setup, use ../batch-prediction/SKILL.md.

Quick workflow

1. Inspect the result directory before changing anything:

   python scripts/inspect_colabfold_outputs.py results_dir
   

2. If outputs are complete enough, interpret scores and plots with references/output-reference.md.
3. If relaxation is requested, choose CPU first unless OpenMM GPU/CUDA is known to work; see references/cli-reference.md.
4. If files or metrics are missing, diagnose owned failures with references/troubleshooting.md before re-running prediction.

Bundled references

• references/cli-reference.md covers colabfold_relax, relax_me(...), plotting/display APIs, extra pTM APIs, and citation writing.
• references/output-reference.md explains expected files, score keys, confidence semantics, validation checks, and safe interpretation.
• references/troubleshooting.md covers optional dependency, backend, data/config, CLI/API, plotting, citation, and workflow failures.
• scripts/inspect_colabfold_outputs.py summarizes PDB/mmCIF/JSON/PNG/BibTeX outputs, detects common missing artifacts, and never downloads, deletes, or rewrites files.