skills

 

Reusable agent skills for Claude Code, Codex, and Antigravity.

Available skills | Install | Contributing | Security

Each skill is a self-contained SKILL.md package with focused activation rules, supporting resources, and validation appropriate to the work it performs. The goal is not to collect prompts. The goal is to make difficult work repeatable.

Available skills

More skills will be added only when they solve a real workflow and include evidence that they work.

Install a skill

These skills use the portable Agent Skills layout. Install a skill directory in the location used by your agent:

Claude Code also supports project skills in .claude/skills/. Codex and Antigravity support repository or workspace skills in .agents/skills/.

Install directly with Codex

Ask Codex to install the latest version:

Use $skill-installer to install the build-open-source-software skill from https://github.com/satwiksps/skills/tree/main/skills/build-open-source-software

For a reproducible install, replace main with a release tag such as v0.1.0.

For the job-search skill, ask Codex:

Use $skill-installer to install https://github.com/satwiksps/skills/tree/main/skills/india-software-jobs

Then open a private tracker folder and run $india-software-jobs /setup. Edit me.md with your preferences and use /scrap, /validate or /repair through the skill. See its short guide for host-specific shortcuts and access requirements.

Install manually with any supported agent

Clone the collection, then copy the skill directory you want into the appropriate location above:

git clone --depth 1 https://github.com/satwiksps/skills.git satwiksps-skills

The commands below install build-open-source-software. Substitute india-software-jobs to install the job tracker. Run the pair for your agent.

macOS and Linux:

# Claude Code
mkdir -p ~/.claude/skills
cp -R satwiksps-skills/skills/build-open-source-software ~/.claude/skills/

# Codex
mkdir -p ~/.agents/skills
cp -R satwiksps-skills/skills/build-open-source-software ~/.agents/skills/

# Antigravity
mkdir -p ~/.gemini/config/skills
cp -R satwiksps-skills/skills/build-open-source-software ~/.gemini/config/skills/

Windows PowerShell:

# Claude Code
New-Item -ItemType Directory -Force "$env:USERPROFILE\.claude\skills" | Out-Null
Copy-Item ".\satwiksps-skills\skills\build-open-source-software" "$env:USERPROFILE\.claude\skills" -Recurse

# Codex
New-Item -ItemType Directory -Force "$env:USERPROFILE\.agents\skills" | Out-Null
Copy-Item ".\satwiksps-skills\skills\build-open-source-software" "$env:USERPROFILE\.agents\skills" -Recurse

# Antigravity
New-Item -ItemType Directory -Force "$env:USERPROFILE\.gemini\config\skills" | Out-Null
Copy-Item ".\satwiksps-skills\skills\build-open-source-software" "$env:USERPROFILE\.gemini\config\skills" -Recurse

Invoke the skill

Use the form supported by your agent:

Claude Code: /build-open-source-software /path/to/idea.md
Codex: Use $build-open-source-software with /path/to/idea.md
Antigravity: Use the build-open-source-software skill with /path/to/idea.md

The shared skill format works across all three agents. Host tools, permissions, and invocation behavior can differ.

Why trust these skills?

The first skill includes a suite of 90 unit and adversarial tests for its deterministic audit tooling. Its checks cover private planning data, transformed content, Git history, authored prose, archive structure, release metadata, legal files, artifact staging, signing contracts, and clean-checkout behavior.

These tests validate the bundled deterministic tooling and repository package. They are not a host-level end-to-end test matrix for every agent UI.

The India job skill tests its local setup and list updates, including duplicate aliases, Applied-mark preservation and evidence-gated closure formatting. Live research depends on the host's web/browser tools; the source seed is not a guarantee of current portal access or exhaustive coverage.

Every skill in this repository must:

• state exactly when it should and should not run;
• keep private inputs and credentials outside public state;
• distinguish verified behavior from claims;
• include a concrete validation path;
• avoid placeholders, fabricated proof, and hidden manual steps;
• document its limitations honestly.

Repository layout

skills/
`-- <skill-name>/
    |-- SKILL.md
    |-- agents/
    |-- assets/
    |-- references/
    |-- scripts/
    `-- tests/

Not every skill needs every optional directory. Every skill does need a valid SKILL.md and enough evidence for its risk level.

Contributing

Bug reports, focused improvements, and real usage reports are welcome. Read CONTRIBUTING.md before proposing a change. If a skill helped you ship something public, use the project showcase issue form so the repository can link only to real work.

Security reports belong in GitHub's private vulnerability reporting flow, not in public issues. See SECURITY.md.

License

Apache License 2.0. See LICENSE and NOTICE.