Skip to main content

Working with Antigravity/Gemini

note

Seqera Cloud users receive $20 in free credits to get started with Co-Scientist. Contact us for additional credits.

The seqera skill command installs a skill file that enables Antigravity/Gemini to use Co-Scientist as a subagent. Once installed, Antigravity can invoke Co-Scientist directly to manage workflows, build containers, query nf-core modules, and more without leaving your environment.

Antigravity/Gemini skill format​

Antigravity/Gemini discovers skills from the .agents/skills/ directory at the repository root. Each skill is a folder containing a SKILL.md file with YAML frontmatter (name, description) and detailed instructions.

AgentFormat
Antigravity/Gemini.agents/skills/

seqera skill install​

Install to the Antigravity skill directory:

seqera skill install --path .agents/skills/seqera-ai-subagent/

Install into the current repository root:

seqera skill install --local

Or install to your home directory:

seqera skill install --global

You can also auto-detect and update an existing installation:

seqera skill install --detect
note

If you encounter a ENOENT: no such file or directory, scandir '/$bunfs/root/content/seqera' error with seqera skill install, you can manually create the skill file. See Manual installation below.

Manual installation​

If the automated installer does not support your agent platform, you can manually create the skill file:

  1. Create the skill directory:

    mkdir -p .agents/skills/seqera-ai-subagent/

  2. Create .agents/skills/seqera-ai-subagent/SKILL.md with the following content:

    ---
    name: seqera-ai-subagent
    description: Invokes Co-Scientist as a domain-expert subagent for Nextflow
    pipeline development, nf-core module management, Seqera Platform workspace
    operations, and Wave container builds.
    ---

    # Co-Scientist Subagent

    When the user asks about Nextflow pipelines, nf-core modules, Seqera Platform,
    or Wave containers, invoke Co-Scientist:

    seqera ai --headless --approval-mode basic "<query>" 2>&1

  3. Verify the installation:

    seqera skill check

Invocation patterns​

Antigravity invokes Co-Scientist dynamically via shell commands rather than static context injection. The recommended patterns are:

PatternCommandUse case
Headless queryseqera ai --headless --approval-mode basic "<query>"Read-only questions, analysis
Sub-agent modeseqera ai --sub-agent --approval-mode basic "<query>"Structured JSONL output
Goal modeseqera ai --headless --approval-mode full "/goal <task>"Multi-step autonomous work
Module QA reviewseqera ai --headless --approval-mode basic "Review modules/nf-core/<module>/main.nf for correctness"Pre-push nf-core module validation

Validated use case: nf-core module QA​

Antigravity uses Co-Scientist as a domain-expert QA gate before pushing nf-core module PRs. In PR #11377 (emmtyper), Co-Scientist caught that emmtyper --version | sed was fragile across Docker/Conda environments due to Click version differences, and recommended using python -c "import emmtyper; print(emmtyper.__version__)" instead.

seqera ai --headless --approval-mode basic \
  "Review modules/nf-core/emmtyper/main.nf for topic channel, stub, and eval correctness" 2>&1

This pattern complements nf-core modules lint by catching semantic issues that static linting misses.

Usage​

seqera skill install [OPTIONS]

Options​

OptionShortDescription
--local-lInstall to repo root
--path <PATH>-pInstall to a custom path (relative or absolute)
--global-gInstall to home directory
--detect-dAuto-detect an existing installation and update it

seqera skill check​

Verify that your installed skill matches your current CLI version:

seqera skill check

Update automatically if needed:

seqera skill check --update

Usage​

seqera skill check [OPTIONS]

Options​

OptionShortDescription
--update-uAutomatically update outdated skills
--globalCheck only global installations
--localCheck only local (repository) installations

Learn more​

On this Page