# AI use cases for Wikiskill (ashutoshsinghpr7/wikiskill) Repo: [ashutoshsinghpr7/wikiskill](https://github.com/ashutoshsinghpr7/wikiskill) | Use cases: 5 | Page: https://dedupedai.com/use-cases/ashutoshsinghpr7-wikiskill/ WikiSkill is a Python framework that lets an AI agent turn its past runs into a persistent wiki and evolve its own skills, keeping only changes that score better on held-out tasks. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Evolve an agent's skills from its own task traces: https://dedupedai.com/use-cases/ashutoshsinghpr7-wikiskill/evolve-agent-skills-overnight/ - Test whether an evolved skill actually improved results: https://dedupedai.com/use-cases/ashutoshsinghpr7-wikiskill/check-if-a-skill-really-helped/ - Run the WikiSkill loop with Claude Code as the agent: https://dedupedai.com/use-cases/ashutoshsinghpr7-wikiskill/run-wikiskill-loop-on-claude-code/ - Evolve skills on a custom task set you write: https://dedupedai.com/use-cases/ashutoshsinghpr7-wikiskill/evolve-skills-on-your-own-tasks/ - Install the distilled Hermes skills from the WikiSkill tap: https://dedupedai.com/use-cases/ashutoshsinghpr7-wikiskill/install-distilled-patterns-into-hermes/ ## How to evolve an agent's skills from its own task traces with Wikiskill Repo: [ashutoshsinghpr7/wikiskill](https://github.com/ashutoshsinghpr7/wikiskill) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Run the full WikiSkill loop so the agent distills failures into a wiki and gates skill changes. **Who it is for:** A developer whose coding agent repeatedly fails the same kinds of tasks and who wants its skills improved without rewriting them by hand. **The problem:** Agent lessons usually disappear when a session ends, and skill edits made by hand are hard to verify. **The outcome:** A workspace with a persistent wiki of pattern pages and a skill set where each kept change beat the best validation score so far. ### Steps 1. Run `pip install wikiskill` to install the package from PyPI. 2. Run `wikiskill init demo` to create the workspace and the 22-task demo bench (13 train, 9 validation). 3. Run `wikiskill status` to see the baseline scores, skills, wiki and history. 4. Run `wikiskill evolve --iters 3` to start the full Algorithm 1 loop with your default model. 5. Open `workspaces/demo/wiki/` and read `index.md` and `skill-impact.md` to see what was learned and which proposals were rejected. ### Prompt to paste into your agent ```text Use the ashutoshsinghpr7/wikiskill repo to create a demo workspace with `wikiskill init demo`, then run `wikiskill evolve --iters 3`. Report the baseline score from `wikiskill status` and whether each proposal was kept or rolled back. Verify that the wiki folder still contains its pattern pages after the run. ``` ### Caveat The README reports that the demo runs are on a small bundled bench, so gains on your own agent may differ. 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/ashutoshsinghpr7-wikiskill/evolve-agent-skills-overnight/ ## How to test whether an evolved skill actually improved results with Wikiskill Repo: [ashutoshsinghpr7/wikiskill](https://github.com/ashutoshsinghpr7/wikiskill) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Use paired per-task comparison with an exact binomial p-value to judge a skill change. **Who it is for:** A team lead or researcher who has an evolved skill and needs evidence it helps before shipping it to other agents. **The problem:** A single average score can hide skills that help on some tasks and hurt on others. **The outcome:** A per-task win, loss and tie count with a two-sided exact-binomial p-value showing whether the skill's effect is distinguishable from chance. ### Steps 1. Run `pip install wikiskill` if the package is not installed yet. 2. Run `wikiskill init demo` and then `wikiskill evolve --iters 3` to produce evolved skills. 3. Run `wikiskill compare` to get the paired statistical comparison. 4. Read the win, loss and tie counts and the p-value in the output. 5. Check docs/COMPARING.md in the repo for how to interpret the result. ### Prompt to paste into your agent ```text In the ashutoshsinghpr7/wikiskill repo, run `wikiskill compare` on the demo workspace after an evolve run. Report the per-task wins, losses, ties and the two-sided exact-binomial p-value. Verify that the comparison used the same validation tasks for both the baseline and the evolved skills. ``` ### Caveat The compare command needs an evolve run first, so the result only covers the tasks in the bundled or your own bench. 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/ashutoshsinghpr7-wikiskill/check-if-a-skill-really-helped/ ## How to run the WikiSkill loop with Claude Code as the agent with Wikiskill Repo: [ashutoshsinghpr7/wikiskill](https://github.com/ashutoshsinghpr7/wikiskill) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Pin a workspace to the Claude backend and evolve skills using Claude Code runs. **Who it is for:** A developer who already uses Claude Code and wants the same evolution loop on that agent instead of Hermes. **The problem:** Teams that standardize on Claude Code want skill evolution in the tool they already use, not a second agent stack. **The outcome:** A workspace pinned to the claude backend, with evolution runs using Claude Code and an isolated CLAUDE_CONFIG_DIR. ### Steps 1. Run `pip install wikiskill` to install the package. 2. Run `claude auth login` once to sign in to Claude Code. 3. Run `wikiskill init demo --backend claude` to create a workspace pinned to Claude Code. 4. Run `wikiskill evolve --iters 3` to run the loop on the pinned backend. 5. Run `wikiskill status` to confirm the scores and the backend used. ### Prompt to paste into your agent ```text Using the ashutoshsinghpr7/wikiskill repo, run `wikiskill init demo --backend claude`, then `wikiskill evolve --iters 2`. Report the baseline and final validation scores and the backend in workspaces/demo/workspace.json. Verify that the workspace file names the claude backend before the evolve run starts. ``` ### Caveat The README says the claude backend needs `claude auth login` once, so the run will not start without a signed-in Claude Code install. 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/ashutoshsinghpr7-wikiskill/run-wikiskill-loop-on-claude-code/ ## How to evolve skills on a custom task set you write with Wikiskill Repo: [ashutoshsinghpr7/wikiskill](https://github.com/ashutoshsinghpr7/wikiskill) | Difficulty: Advanced | Time: 1 hour | Basis: From the README Write tasks.json with auto-graded checks and run the loop on your own agent work. **Who it is for:** A developer with a repeatable agent job, such as formatting data to a spec, who needs the loop to learn from their own examples. **The problem:** The bundled demo bench does not cover your team's tasks, so skill evolution has nothing relevant to learn from. **The outcome:** A custom bench where each task has an automatic grader, and an evolve run that gates skill changes against your validation tasks. ### Steps 1. Write a `tasks.json` file where each task has an id, a split of train or val, a prompt, a sandbox and a grader. 2. Choose a grader type from exact, contains, json_field or code_stdout that fits each task's output. 3. Run `wikiskill init ` to create a workspace, then place your tasks in its bench/tasks folder (inferred). (inferred, check the README) 4. Run `wikiskill status` to check the workspace loaded the tasks. 5. Run `wikiskill evolve --iters 3` to evolve skills on your tasks. ### Prompt to paste into your agent ```text In the ashutoshsinghpr7/wikiskill repo, add my tasks.json with train and val splits and at least one task per grader type I use. Run `wikiskill evolve --iters 2` and report the baseline score and any rejected proposals. Verify that every task has a grader and that missing deliverables score 0 rather than crashing. ``` ### Caveat The README gives the tasks.json format only as an example, so you must check task ids and splits carefully before a long run. 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/ashutoshsinghpr7-wikiskill/evolve-skills-on-your-own-tasks/ ## How to install the distilled Hermes skills from the WikiSkill tap with Wikiskill Repo: [ashutoshsinghpr7/wikiskill](https://github.com/ashutoshsinghpr7/wikiskill) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Add the WikiSkill skills tap to Hermes and install the patterns from live runs. **Who it is for:** A Hermes Agent user who wants proven fixes, such as the ripgrep binary-skip pitfall, without running an evolution loop first. **The problem:** Hermes agents hit the same sandbox and search pitfalls again and again, and writing fixes from scratch takes time. **The outcome:** Hermes has the selected WikiSkill skills installed and ready to use in your agent sessions. ### Steps 1. Run `hermes skills tap add ashutoshsinghpr7/wikiskill` to register the tap. 2. Run `hermes skills install ashutoshsinghpr7/wikiskill/skills/search-miss-binary` to install the search pattern. 3. Run `hermes skills install ashutoshsinghpr7/wikiskill/skills/script-exec-blocked` to install the sandbox policy pattern. (inferred, check the README) 4. Repeat the install command for each other skill you want from the table in the README. 5. Start a Hermes session and confirm the installed skills appear in the skill list. ### Prompt to paste into your agent ```text Using the ashutoshsinghpr7/wikiskill repo as a Hermes skills tap, add the tap and install the skills search-miss-binary and script-exec-blocked. Report which skills were installed. Verify that each skill name appears in the Hermes skill list after the install. ``` ### Caveat The README lists these skills as patterns from its own demo runs, so they may not match the failures in your environment. 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/ashutoshsinghpr7-wikiskill/install-distilled-patterns-into-hermes/