Skills

Fastfold skills are packaged instructions and scripts that help AI agents run common workflows end-to-end.

Built on the open Agent Skills ecosystem from Vercel, they work across many agent tools, including the Fastfold Agent CLI, Claude Code, Cursor, Codex, GitHub Copilot, Gemini, and others. See skills.sh for the current list of supported agents.

Skill directories and name: frontmatter use snake_case (for example md_openmm_calvados, protein_design_boltzgen). Install with the same name or the owner/repo@skills/<name> subpath.

Install

npx skills add fastfold-ai/skills                  # all skills
npx skills add fastfold-ai/skills --skill fold     # one skill

In interactive Agent CLI sessions, use /skills, /skills-find, /skills-add, /skills-upgrade, and /skills-remove. The agent can also discover and install skills when you enable fastfold config set skills.allow_agent_install true.

fastfold setup includes a skills step: it live-fetches the Fastfold catalog (all skills preselected), suggests community collections, and accepts custom sources. Non-interactive: fastfold setup --skills "fastfold-ai/skills@skills/fold,..." or --skip-skills.

Repository: github.com/fastfold-ai/skills
Listing: skills.sh/fastfold-ai/skills
Agent CLI skill management: Agent CLI guide

Install method and location (Agent CLI)

• When Node/npx is available, installs prefer npx skills add into a Fastfold-owned dir (~/.fastfold-cli/.claude/skills).
• Otherwise a native git clone installs into ~/.fastfold-cli/skills.
• Both locations are discovered automatically from any working directory.

Versioning and updates (Agent CLI)

The Fastfold catalog is published with GitHub Releases (one semver tag per catalog version), and the CLI surfaces that release as a version.

• fastfold skills list (and /skills) show Author, Updated, and Version. The version is the catalog's latest release tag (for example v1.0.0) and the author is the source repo org (for example fastfold-ai). Sources without a release — npx-installed or bundled helper skills — show Version not available.
• Run fastfold skills upgrade (or /skills-upgrade) to sync the catalog and update installed skills to the latest release.
• On launch the CLI shows a non-blocking notice when a newer catalog release is available (Skills update available: v1.0.0 -> v1.1.0 Run /skills-upgrade). The check reads a local cache and refreshes in the background, so it never slows startup.

Community skill collections

Install whole collections from other providers:

fastfold skills add K-Dense-AI/scientific-agent-skills   # K-Dense-AI
fastfold skills add anthropics/life-sciences             # Anthropic
fastfold skills add google-deepmind/science-skills       # DeepMind

The catalog grows over time — run fastfold skills find (or /skills-find) to see what is available now.

Bundled vs installable (Agent CLI)

Only two helper skills ship bundled with the Agent CLI:

Domain skills (boltz, fold, md_openmm_calvados, md_openmmdl, protein_design_boltzgen, slack_report, and future catalog additions) are installable, not bundled. Install them from the Fastfold catalog before use.

Running skill scripts

Skill scripts are portable and self-contained. From a skill's directory:

python scripts/<script_name>.py <args>

In the Agent CLI, the installed-skills section of the system prompt shows each skill's absolute directory and rewrites script paths so the agent can invoke them correctly.

Available skills (Fastfold catalog)

boltz

Direct Boltz API automation that drives the official boltz-api CLI for structure-and-binding, protein/small-molecule design and screen, ADME, and status/recovery.

Use it when you need:

• direct boltz-api execution instead of Fastfold Jobs/Workflows wrappers
• an estimate -> confirm -> submit -> wait/download flow (it always estimates and waits for your approval before any billable run)
• durable artifacts and API-based recovery: locally it writes into the agent output dir (${OUTPUT_DIR:-./outputs}/boltz); in a hosted sandbox it downloads to /tmp and persists to the S3-backed /workspace with scripts/persist.sh

Requires BOLTZ_API_KEY (see API key setup).

Install:

fastfold skills add fastfold-ai/skills@skills/boltz

References:

• Skill: skills/boltz/SKILL.md
• Persistence helper: skills/boltz/scripts/persist.sh
• API mapping: skills/boltz/references/api.md
• Results/runtime behavior: skills/boltz/references/results.md
• Quick examples: skills/boltz/references/examples.md

Example prompts:

Run a simple ROR1-style Boltz-2 structure-and-binding smoke test with aspirin; estimate first, then execute.

Run a minimal AMBP-style protein design job with 10 proteins and return top metrics and artifact paths.

Screen aspirin, ibuprofen, and caffeine against a PknB-style target with small-molecule library screen and summarize top hits.

Run ADME quick triage for aspirin, ibuprofen, phenol, and caffeine and report lipophilicity/permeability/solubility.

Recover this Boltz job by ID (retrieve) and resume local artifacts with run name sms-pknb-demo.

fold

Fastfold Jobs API automation for protein folding.

Use it when you need:

• protein sequence folding with Fastfold
• API or script-based job orchestration
• automated create -> wait -> fetch pipelines

Install:

fastfold skills add fastfold-ai/skills@skills/fold

References:

• Skill: skills/fold/SKILL.md
• Auth/API: skills/fold/references/auth_and_api.md
• Schema summary: skills/fold/references/schema_summary.md
• OpenAPI reference: skills/fold/references/jobs.yaml

md_openmm_calvados

CALVADOS + OpenMM workflow automation via Workflows API (calvados_openmm_v1).

Use it when you need:

• fold -> OpenMM linked simulation flows
• manual PDB + PAE submission for MD simulation
• MD metrics, artifacts, and result polling

Install:

fastfold skills add fastfold-ai/skills@skills/md_openmm_calvados

References:

• Skill: skills/md_openmm_calvados/SKILL.md
• Auth/API: skills/md_openmm_calvados/references/auth_and_api.md
• Schema summary: skills/md_openmm_calvados/references/schema_summary.md
• Method details: skills/md_openmm_calvados/references/calvados_method.md

md_openmmdl

OpenMMDL protein-ligand molecular dynamics workflow automation via Workflows API (openmmdl_v1).

Use it when you need:

• topology + ligand-driven MD workflow submission
• draft script preparation + execution for OpenMMDL
• analysis outputs, trajectory artifacts, and frame extraction

Install:

fastfold skills add fastfold-ai/skills@skills/md_openmmdl

References:

• Skill: skills/md_openmmdl/SKILL.md
• Auth/API: skills/md_openmmdl/references/auth_and_api.md
• Schema summary: skills/md_openmmdl/references/schema_summary.md

protein_design_boltzgen

BoltzGen protein design workflow automation (draft -> graph upsert -> review -> execute -> ranked results).

Use it when you need:

• single-spec or multi-spec BoltzGen design flows
• example-first workflow setup from bundled local preset files
• candidate metrics interpretation and per-candidate Mol* result links
• API-only workflow orchestration that mirrors Composer behavior

Install:

fastfold skills add fastfold-ai/skills@skills/protein_design_boltzgen

References:

• Skill: skills/protein_design_boltzgen/SKILL.md
• Workflow helper script: skills/protein_design_boltzgen/scripts/workflow_api.py
• YAML keys: skills/protein_design_boltzgen/references/yaml_keys.md
• Preset examples: skills/protein_design_boltzgen/references/examples/README.md
• Metrics guide: skills/protein_design_boltzgen/references/metrics_guide.md

slack_report

Share CLI markdown reports to Slack and save a library copy in Fastfold Cloud.

Use it when you need:

• share a markdown report to Slack
• keep a saved copy in Fastfold library
• give teammates a quick link to open the saved report

Install:

fastfold skills add fastfold-ai/skills@skills/slack_report

References:

• Skill: skills/slack_report/SKILL.md
• API reference: skills/slack_report/references/api.md

Example prompts

After installing, ask your agent directly:

Use Boltz-2 in Fastfold with affinity property to the ligand. Fold this protein: PQITLWQRPLVTIKIGGQLKEALLDTGADDTVLEEMSLPGRWKPKMIGGIGGFIKVRQYDQILIEICGHKAIGTVLVGPTPVNIIGRNLLTQIGCTLNF and this ligand: CC1CN(CC(C1)NC(=O)C2=CC=CC=C2N)C(=O)NC(C)(C)C

Use Boltz API directly to run structure-and-binding for this target+ligand payload, estimate cost first, then execute and return persistent artifact paths.

Run an OpenMM simulation from my completed fold job and return metrics, plots, and the dashboard link.

Run an OpenMMDL simulation with topology + ligand inputs and return analysis outputs with links.

Help me run a quick BoltzGen smoke test for a simple peptide binder against 5CQG, and show me the draft before running.

Show me BoltzGen protein design examples and run vanilla_target_binding_site with one target and one budget in the background. Keep the workflow_id for follow-up results.

Share the current report to Slack and save a copy in my Fastfold library.

API key setup

These skills run against two compute providers. Set the key for whichever skills you use.

Option A: .env (recommended)

Your agent can help you create these files and follow the setup.

cp skills/fold/references/.env.example .env

Then edit .env, setting only the key(s) for the providers you use:

FASTFOLD_API_KEY=sk-your-actual-key-here
BOLTZ_API_KEY=sk-your-boltz-key-here

Option B: shell environment

export FASTFOLD_API_KEY="sk-..."
export BOLTZ_API_KEY="sk-..."

Environment variables take precedence over .env. In a hosted sandbox, set the variable and restart the session so it becomes visible.

Slack setup for slack_report

Before using /export-share, connect Slack and set a channel in Fastfold Cloud:

• cloud.fastfold.ai/integrations/slack

Set or confirm the channel for the agent_cli_report mode.

Security note

Keep API keys local and never commit .env files.