Struct Predictor

What it does

When your agent uses it

• Tasks that involve Protein structure and design

Example prompts

• “/struct-predictor”

Requirements

• Python 3

Workflow steps

4 steps, taken from the first numbered list in SKILL.md.

1. 1Structure Prediction: Run Boltz-2 (default) or OpenFold3 (--backend openfold3, CUDA GPU required) locally on a YAML input
2. 2Confidence Extraction: Per-residue pLDDT (from CIF B-factors) and PAE matrix (from confidence JSON)
3. 3Report Generation: Markdown with pLDDT line plot, PAE heatmap, band breakdown, and reproducibility bundle
4. 4Demo Mode: Trp-cage miniprotein (20 residues, PDB 1L2Y) — runs immediately, no input required

What it can do on your machine

Read from SKILL.md and the folder at commit dece754. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Ships script files (Python), which the agent can run.

  Shell commands in SKILL.md call:

  • uv
  • python
  • pip

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Hosts in commands or code, which the agent is likely to contact:

  • download.pytorch.org
  • pypi.org

  Also links to:

  • github.com

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Struct Predictor loads about 1.8k tokens when it runs. Until then it costs about 67 tokens; SKILL.md has 496 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~67

When it runs · the whole SKILL.md, loaded when a task matches

~1.8k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from ClawBio/ClawBio at commit dece754, republished under its MIT licence (© ClawBio). 496 words, ~1,849 tokens.

name

struct-predictor

description

Protein structure prediction with Boltz-2 (default) or OpenFold3. Accepts YAML inputs (single protein or multi-chain complex), runs the chosen backend offline, extracts per-residue pLDDT and PAE confidence, and writes a markdown report with figures.

license

MIT

metadata.version

0.3.0

# Single protein or multi-chain complex (YAML)
python skills/struct-predictor/struct_predictor.py \
  --input complex.yaml --output /tmp/struct_out

# Same input with OpenFold3 instead of the Boltz-2 default
python skills/struct-predictor/struct_predictor.py \
  --input complex.yaml --output /tmp/struct_out --backend openfold3

# Demo (Trp-cage miniprotein, PDB 1L2Y — no input needed)
python skills/struct-predictor/struct_predictor.py \
  --demo --output /tmp/struct_demo

python skills/struct-predictor/struct_predictor.py --input my_protein.yaml --output /tmp/struct_out

python skills/struct-predictor/struct_predictor.py --demo --output /tmp/struct_demo

python skills/struct-predictor/struct_predictor.py --input complex_ab.yaml --output /tmp/complex_out

output_dir/
  predictions/[name]/                      # Boltz native output (default backend)
    [name]_model_0.cif                     # predicted structure (pLDDT in B-factors)
    confidence_[name]_model_0.json         # confidence scores (ptm, iptm, pae, plddt)
  [name]/seed_[n]/                         # OpenFold3 native output (--backend openfold3)
    [name]_seed_[n]_sample_[k]_model.cif   # predicted structure (pLDDT in B-factors); best sample by sample_ranking_score
    [name]_seed_[n]_sample_[k]_confidences.json             # per-atom plddt, pae, pde
    [name]_seed_[n]_sample_[k]_confidences_aggregated.json  # avg_plddt, ptm, iptm, sample_ranking_score
  report.md                                # primary markdown report
  viewer.html                              # self-contained 3Dmol.js 3D viewer (open in browser)
  result.json                              # machine-readable summary
  figures/
    plddt.png                              # per-residue pLDDT confidence plot
    pae.png                                # PAE inter-residue error heatmap
  reproducibility/
    commands.sh                            # exact struct_predictor.py command used
    environment.txt                        # backend package version snapshot

version: 1
sequences:
  - protein:
      id: A
      sequence: ACDEFGHIKLMNPQRSTVWY
      msa: empty        # runs offline; replace with a path to a .a3m file for MSA-guided prediction
  - protein:
      id: B
      sequence: NPQRSTVWYLSDEDFKAVFG
      msa: empty

uv pip install boltz -U          # CPU
uv pip install "boltz[cuda]" -U  # GPU (recommended)
uv pip install openfold3         # optional, --backend openfold3; CUDA GPU required
setup_openfold                   # once: downloads OpenFold3 weights (~2 GB)
uv pip install numpy matplotlib pyyaml

© ClawBio, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 10 other files in skills/struct-predictor of ClawBio/ClawBio.

• SKILL.md
• demo_data/trpcage.yaml
• struct_predictor.py
• struct_predictor_core/__init__.py
• struct_predictor_core/confidence.py
• struct_predictor_core/io.py
• struct_predictor_core/predict.py
• struct_predictor_core/report.py
• struct_predictor_core/viewer.py
• tests/__init__.py
• tests/test_struct_predictor.py

Open the folder on GitHubat commit dece754