Whalory Core, free.
One skill folder for the assistants that read Agent Skills. Text under CC BY 4.0, scripts under MIT.
Install Whalory 3.2.1
Whalory Core is free. Download and extract the install kit, then choose the tool you use.
Download the free install kitDocumentation checked
Extract the kit into your project folder and run this command there. It installs Whalory for that project. Some tools also need the manual setup steps in their documentation.
py -3 .\install\install.py install --host codex --scope project --version 3.2.1 --edition core --project-root (Get-Location).PathAbout this install route
This guide has been checked against the tool’s official documentation. This version has not yet been tested inside the tool.
Guide last checked: October 4, 2026
Individual downloads and file checksums
To check that a download is intact, compare its checksum with the value below.
whalory-core-3.2.1-claudeai.zip
1.2 MB
87e5919e98d1c9868a7c113a69af89c04f78a16e6efb2c9aaf3cab5f6c0ec987
whalory-core-3.2.1-claude-plugin.zip
1.4 MB
e4e82281f54b0c484a65f41a83a3f001d7522f52c39aec85058e925e3ab11850
whalory-core-3.2.1-agent-plugin.zip
1.3 MB
f2547f11662e82df9ecaddda3411abaecb08b734e600b2efb9a7a94a034f6c70
whalory-core-3.2.1-codex-plugin.zip
1.3 MB
86e616dfe6c2237b1b06101c4a507ec263486e7a4b225f1c783223b14842300d
whalory-core-3.2.1-cursor-plugin.zip
1.3 MB
a9bd6a82014f77eb6455931a1bd8d3c75c1f4fa032c1f70b0640aae68a03136e
whalory-core-3.2.1-gemini-ext.zip
1.3 MB
ee55787ebd6bd568b7700a9c1fa8cb68e2f5b3101d004af4963e506f179ab91a
whalory-core-3.2.1-offline-kit.zip
10.1 MB
30035624625d135ff569c41d781e7c2f829a6b6f92422d516b625a809437e0f0
The Core skill folder
A zip file with the whalory folder, 1.2 MB, 197 files.
whalory-core-3.2.0-skill.zip
0a357ed1e2535e3ad22c181f9a7277bd4f986cbcb91fe9d4ea7475c9ee536801To check the file after downloading:
shasum -a 256 whalory-core-3.2.0-skill.zipGet-FileHash whalory-core-3.2.0-skill.zip -Algorithm SHA256Source code and release history are available in the public repository. Check the version before choosing a package. See the repository
What is inside
- SKILL.md, the map every assistant reads first.
- Playbooks for the most common tasks, and craft files in English and Persian.
- Voice profile templates and starter profiles.
- Linters, self-tests, and an MCP server with six tools, for Python 3.8 or later.
Python is optional. Without it, Whalory reviews with a manual checklist and says “QA: manual” in its note.
Where will you use it?
Pick your assistant for the exact steps. Each guide was checked against that tool’s documentation on the date it shows.
We say a tool works only after a dated test. Until then a guide says “checked against the documentation”.
Install guides
- Install Whalory Core in Claude Code
- Install Whalory Core in Codex
- Install Whalory Core in Cursor
- Install Whalory Core for GitHub Copilot in VS Code
- Install Whalory Core in Gemini CLI
Check that it works
Either one to three short questions arrive, each with a recommended option, or the caption arrives with a note that starts with “Diagnosis:”. Unknown facts stay in brackets.
Every route, column by column.
Six columns: instructions, executable checks, independent review, a stored profile, Hub participation, and updates. A pasted prompt brings none of the local tools, memory, separate agents, or automatic updates.
14 install routes shown.
- Available
- Released in 3.0.0 and reachable today.
- Limited
- Usable, with the restriction written next to it.
- Planned
- Not in 3.0.0.
- Not supported
- This route cannot do it.
Claude Code plugin
AvailableThe Core plugin is available from the download page and the public GitHub release. The Pro plugin is available with a license.
What this route includes
- InstructionsAvailable
- The full skill plus the write, review, voice, and lint commands; Pro adds campaign, transcreate, and audit.
- Executable checksAvailable
- Local MCP tools, six in Core and ten in Pro. On Windows, install with --config python=python.
- Independent reviewAvailable
- A separate editor subagent reviews each draft blind. It gets the draft, the facts, the profile, and the format, never the writer’s reasoning.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Download and replace the plugin to update it manually. Marketplace updates have not been tested end to end. The private Pro marketplace on whalory.com requires Claude Code 2.1.238 or later.
Test record: During packaging, Claude Code 2.1.278 loaded the plugin with --plugin-dir and listed its MCP server as connected. The marketplace install has not been run.
Claude Code skill folder
AvailableThe Core skill folder is a free download today; the install path is checked against Claude Code’s documentation.
Open the install guideWhat this route includes
- InstructionsAvailable
- The whole method in one folder, called with /whalory or by describing the task.
- Executable checksAvailable
- The checking scripts run in Claude Code with Python 3.8 or later. Without Python the note says “QA: manual”.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual: replace the whalory folder with the new version.
Test record: No end-to-end test on this host is recorded.
claude.ai and Claude Desktop skill upload
LimitedCode execution must be enabled in Claude. Download the Core claude.ai package from the download page; Pro includes its own package.
What this route includes
- InstructionsAvailable
- Upload the -claudeai.zip in Customize > Skills. Its description fits the 200-character limit; the plain skill zip does not.
- Executable checksLimited
- Scripts can run only in Claude’s code execution sandbox, and they have not been tested there.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileLimited
- No local profile folder: paste VOICE.md into the chat or keep it in a Claude project.
- Hub participationNot supported
- The Hub needs a local install with Python; this route has none.
- Update methodLimited
- Manual: upload the new files again.
Test record: Not yet tested with a real upload.
Claude Desktop extension (MCPB)
LimitedAdds the tools only and needs Python 3.8 or later, which Claude Desktop does not include. The Core bundle is available on the download page.
What this route includes
- InstructionsLimited
- The bundle carries no method: upload the claude.ai skill as well.
- Executable checksAvailable
- Read-only lint and reference tools that make no network calls.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileLimited
- profile_lookup reads VOICE.md only inside the folders you allow.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual: install the new .mcpb file.
Test record: No end-to-end test on this host is recorded.
Custom GPT pack
LimitedPro pack. OpenAI now lets only Business, Enterprise, and Edu workspaces create new GPTs, and plans to retire custom GPTs.
What this route includes
- InstructionsLimited
- instructions.txt stays under 8,000 characters, with 14 knowledge files in the English pack, 16 in the Persian pack, plus VOICE.md.
- Executable checksLimited
- The scripts in extras/ run only when Code Interpreter is on. Not tested.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileLimited
- VOICE.md goes in as a static knowledge file. Nothing is written back to it.
- Hub participationNot supported
- The Hub needs a local install with Python; this route has none.
- Update methodLimited
- Manual: replace the instructions and knowledge files.
Test record: No end-to-end test on this host is recorded.
ChatGPT project instructions
LimitedPaste a ready-made prompt into the project instructions. The Core prompts are available from the download page; Pro includes longer prompts.
What this route includes
- InstructionsLimited
- A 4,000-character prompt in Core, 8,000 in Pro. It carries a summary of the method; the full references stay out.
- Executable checksNot supported
- No scripts run. Whalory checks by hand and writes “QA: manual” in its note.
- Independent reviewLimited
- Self-review in the same chat: the prompt asks for three editing passes and a manual checklist.
- Stored profileLimited
- Add VOICE.md as a project file.
- Hub participationNot supported
- The Hub needs a local install with Python; this route has none.
- Update methodLimited
- Nothing updates by itself: paste the new prompt.
Test record: No end-to-end test on this host is recorded.
Gemini CLI
AvailableInstall the Core skill in Gemini CLI’s skills folder, or download the extension and install it from its extracted local folder. A separate public repository for the extension has not been created.
Open the install guideWhat this route includes
- InstructionsAvailable
- Skill in ~/.gemini/skills/ or ~/.agents/skills/. Gemini activates it when a task matches, after you confirm.
- Executable checksAvailable
- The checking scripts run with Python 3.8 or later; the extension starts its MCP server with python3.
- Independent reviewLimited
- The extension ships writer and editor subagents, a preview feature in Gemini CLI. The plain skill reviews in a separate pass.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual for now: install the new package again.
Test record: No end-to-end test on this host is recorded.
Gem pack
LimitedPro pack for Gems. Google’s help page states no file limit; the pack assumes ten files, based on a 2024 report.
What this route includes
- InstructionsLimited
- Gem instructions plus nine knowledge files; the tenth slot is for VOICE.md.
- Executable checksNot supported
- No scripts run. Whalory checks by hand and writes “QA: manual” in its note.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileLimited
- VOICE.md goes in as a static knowledge file. Nothing is written back to it.
- Hub participationNot supported
- The Hub needs a local install with Python; this route has none.
- Update methodLimited
- Manual: upload the new files again.
Test record: No end-to-end test on this host is recorded.
Gemini app skill upload
LimitedPro upload. Skills work only in Gemini Spark, for adults with a personal Google account, Google AI Pro or Ultra, and Keep Activity on. They are not offered in the European Economic Area, Nigeria, Switzerland, or the UK.
What this route includes
- InstructionsLimited
- Upload the -gemini-app.zip on the Skills page: text files only, 100 MB in total.
- Executable checksLimited
- Whether the scripts run there is not verified.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileLimited
- Paste VOICE.md into the chat.
- Hub participationNot supported
- The Hub needs a local install with Python; this route has none.
- Update methodLimited
- Manual: upload the new files again.
Test record: No end-to-end test on this host is recorded.
Cursor
AvailableThe Core skill folder installs in Cursor’s skills folders today; the steps are checked against Cursor’s documentation. The Pro Cursor plugin comes with a license.
Open the install guideWhat this route includes
- InstructionsAvailable
- Skill in .cursor/skills/ or .agents/skills/ in the project, or in the same folders under ~; type / in Agent chat to call it.
- Executable checksAvailable
- MCP tools through .cursor/mcp.json with Python 3.8 or later. The Agent Plugins build does not start them in Cursor.
- Independent reviewLimited
- The plain skill reviews in a separate pass. The Pro Cursor plugin carries writer and editor agents; not tested.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual: replace the folder. The Pro plugin lives in ~/.cursor/plugins/local/.
Test record: No end-to-end test on this host is recorded.
Codex
AvailableInstall the Core skill in ~/.agents/skills/. The Core Agent Plugins package is also available from the download page; follow the guide for your host.
Open the install guideWhat this route includes
- InstructionsAvailable
- Called with $whalory or picked from /skills.
- Executable checksLimited
- Add the MCP server to ~/.codex/config.toml with --root. From the plugin, codex-cli 0.145.0 shared no project folder, so only lint_text worked.
- Independent reviewLimited
- Commands arrive as skills and Codex gets no separate reviewer agent, so review is a separate pass.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual: replace the whalory folder with the new version.
Test record: codex-cli 0.145.0: the skills loaded; the plugin’s MCP server started only from the legacy .mcp.json and received no project folder.
VS Code and GitHub Copilot
AvailableThe Core skill folder installs in .github/skills/ or ~/.copilot/skills/ today. Neither plugin route has been tested in VS Code.
Open the install guideWhat this route includes
- InstructionsAvailable
- Copilot reads skills in agent mode in VS Code and JetBrains, in the Copilot CLI, and in the cloud agent.
- Executable checksAvailable
- MCP tools through .vscode/mcp.json with Python 3.8 or later.
- Independent reviewLimited
- The Agent Plugins build carries writer and editor agents for Copilot, not yet tested. The plain skill reviews in a separate pass.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual: replace the whalory folder with the new version.
Test record: Neither plugin route had been tested in VS Code when 3.0.0 was built.
Other skill and MCP hosts
LimitedHosts that read Agent Skills from .agents/skills/ can load the folder, and the tools connect by hand as a stdio server. Paths differ by host and are not all verified.
What this route includes
- InstructionsLimited
- The skill folder, where the host reads Agent Skills.
- Executable checksLimited
- Point the host at scripts/mcp_server.py with Python 3.8 or later.
- Independent reviewLimited
- Self-review: the same model rereads the draft in a separate pass. No separate reviewer agent.
- Stored profileAvailable
- Reads VOICE.md at the project root, or ~/.whalory/profiles/, on every task, with LEARNINGS.md beside it.
- Hub participationLimited
- Update files manually. Hub sync and statistics sharing are disabled in this release. Local tools work without them.
- Update methodLimited
- Manual: replace the whalory folder with the new version.
Test record: No end-to-end test on this host is recorded.
Copy and paste
LimitedFor any chat with a custom instructions field. A pasted prompt brings no local tools, no memory, no separate agents, and no updates. The Core prompts are available on the download page.
What this route includes
- InstructionsLimited
- 1,500 or 4,000 characters in Core, 8,000 in Pro: a summary of the method.
- Executable checksNot supported
- No scripts run. Whalory checks by hand and writes “QA: manual” in its note.
- Independent reviewLimited
- Self-review in the same chat: the prompt asks for three editing passes and a manual checklist.
- Stored profileLimited
- No memory: paste VOICE.md at the start of each chat, or fill in the profile block inside the prompt.
- Hub participationNot supported
- The Hub needs a local install with Python; this route has none.
- Update methodLimited
- Nothing updates by itself: paste the new prompt.
Test record: No end-to-end test on this host is recorded.
Product names belong to their makers. Listing them does not mean their makers endorse Whalory.
Other Core packages
Core packages are available on GitHub for supported assistants, plugins, and paste-in prompts. Pick your assistant below to see the package and any setup requirements.
- claude.ai / Claude Desktop upload
whalory-core-3.2.0-claudeai.zipDownload 1.2 MB - Claude Code plugin
whalory-core-3.2.0-claude-plugin.zipDownload 1.4 MB - Agent Plugins package (Codex, Cursor, Copilot)
whalory-core-3.2.0-agent-plugin.zipDownload 1.3 MB - Codex plugin (legacy layout)
whalory-core-3.2.0-codex-plugin.zipDownload 1.3 MB - Cursor plugin
whalory-core-3.2.0-cursor-plugin.zipDownload 1.3 MB - Gemini CLI extension
whalory-core-3.2.0-gemini-ext.zipDownload 1.3 MB - Claude Desktop extension (MCPB)
whalory-core-mcp-3.2.1.mcpbDownload 1.3 MB - Paste-in prompts
whalory-core-3.2.0-paste.zipDownload 28 KB
Pro and Studio packages
With Pro or Studio
Paid packages are never public files. Once you buy, they download from your license page, with links that expire.