# AI use cases for Misaka Agent (Luciole-Studio/Misaka-Agent) Repo: [Luciole-Studio/Misaka-Agent](https://github.com/Luciole-Studio/Misaka-Agent) | Use cases: 3 | Page: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/ MISAKA is a Python multi-agent research system for the humanities and social sciences that turns one question into a reviewed research graph and a sourced research article. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Run a research question across several branches: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/run-a-multi-branch-research-question/ - Check each research claim against its source page: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/trace-claims-to-source-pages/ - Expose the unstated premises in a draft argument: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/surface-unstated-premises-in-an-argument/ ## How to run a research question across several branches with Misaka Agent Repo: [Luciole-Studio/Misaka-Agent](https://github.com/Luciole-Studio/Misaka-Agent) | Difficulty: Intermediate | Time: 60 minutes | Basis: From the README Turn one humanities question into a reviewed research graph and a sourced research article. **Who it is for:** A humanities or social science researcher with a question that has several defensible readings and no time to follow each one by hand. **The problem:** A single model answer tends to settle on the most common reading and leave the other readings unexamined. **The outcome:** A research article in your project folder, with alternative branches recorded as graph nodes and each citation traceable to its file and page. ### Steps 1. Run `uv tool install "misaka[providers] @ git+https://github.com/Luciole-Studio/Misaka-Agent.git"`. 2. Run `mkdir ~/Documents/my-research` and then `cd ~/Documents/my-research`. 3. Run `misaka setup` to sign in, choose a model and create your first two Sisters. 4. Run `misaka`, then type `/research` and answer the parameter prompts, or give them on one line such as `/research --depth 2 --max-nodes 12 QUESTION`. 5. Approve the plan in conversation when Last Order presents it, since no plan proceeds without your approval. 6. Read the Markdown files in the project folder, where the plans, cards, conclusions and source lists are kept. ### Prompt to paste into your agent ```text Use the Luciole-Studio/Misaka-Agent repo to run /research on this question: . Expected output: a research article written into the project folder, with each citation tied to a file and page. Verify that you approved the plan before any Sister began work. ``` ### Caveat Each node uses model usage, and Claude accounts are billed by Anthropic per token as extra usage, so set the depth and node limits before you start a run. Risk, credentials: Setup asks you to sign in with a model account or an API key, so keep those credentials in your own environment and out of shared project files. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/run-a-multi-branch-research-question/ ## How to check each research claim against its source page with Misaka Agent Repo: [Luciole-Studio/Misaka-Agent](https://github.com/Luciole-Studio/Misaka-Agent) | Difficulty: Intermediate | Time: 60 minutes | Basis: From the README Audit every finding in a MISAKA conclusion against the file, page and quotation it names. **Who it is for:** A researcher who must verify each citation in an argument before submitting it to a journal or supervisor. **The problem:** Citations in generated text are often hard to check, so verifying them by hand can take hours. **The outcome:** A list of findings, each with its claim type, source file, page and quotation, plus a version history you can review. ### Steps 1. Run `uv tool install "misaka[providers] @ git+https://github.com/Luciole-Studio/Misaka-Agent.git"`. 2. Place your source files in the project folder before the run, so the Sisters can read them. (inferred, check the README) 3. Run `misaka` and then `/research` with your question, letting the cards and conclusion finish. 4. Open the card files and conclusion versions in the project folder and compare each cited claim with the file and page it names. 5. Commit the project in git only after you have reviewed the files, since MISAKA commits only at your request. ### Prompt to paste into your agent ```text In the Luciole-Studio/Misaka-Agent project folder , open the card and conclusion Markdown files and check that each cited claim matches the file, page and quotation it names. Expected output: a list of citations that match and a list of those that do not. Verify that every finding includes a quotation from its source. ``` ### Caveat The README says citations trace to page and character offset but describes that mechanism in a section not included here, so confirm it on your own material before relying on it for publication. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/trace-claims-to-source-pages/ ## How to expose the unstated premises in a draft argument with Misaka Agent Repo: [Luciole-Studio/Misaka-Agent](https://github.com/Luciole-Studio/Misaka-Agent) | Difficulty: Intermediate | Time: 60 minutes | Basis: From the README Use a red-team review to find what a conclusion leaves unsaid yet relies on. **Who it is for:** A social scientist whose draft argument may depend on assumptions it never states, and who wants them tested before peer review. **The problem:** Co-authors and reviewers often miss the premises a conclusion takes for granted. **The outcome:** A set of recorded gaps and presuppositions, each answered as a revision, rebuttal, conceded cost, referral, parked point or decision for you. ### Steps 1. Run `uv tool install "misaka[providers] @ git+https://github.com/Luciole-Studio/Misaka-Agent.git"`. 2. Run `mkdir ~/Documents/my-research` and then `cd ~/Documents/my-research`. 3. Run `misaka`, then type `/research` with your draft argument stated as the question. 4. Read the red-team review and the divergence review recorded in each node's files. 5. Answer any decision MISAKA raises in conversation, because it seeks your consent before altering the question you asked. ### Prompt to paste into your agent ```text Use the Luciole-Studio/Misaka-Agent repo to run /research on this draft argument: . Expected output: a red-team review and a divergence review listing unstated premises and gaps in the conclusion. Verify that each listed presupposition has a recorded response in the node's files. ``` ### Caveat Each Sister brings her own model, so the strength of the red-team review depends on which model you assign to that Sister. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/luciole-studio-misaka-agent/surface-unstated-premises-in-an-argument/