Skill management
Add reusable working instructions to EvoX, verify sources and dependencies, and manage activation and maintenance. See Skill workflows for task design and outcome validation.
Choose a source
Open Extensions > Skills. In coding mode, /skills also opens management. Import a skill directory or package; a .gepx evolution-history archive is not a skill package.
| Source | Check before adding |
|---|---|
| Local scan | Complete instructions, scripts, and reference files |
| ZIP import | Archive source, directory structure, and bundled scripts |
| GitHub import | Correct repository and path; read access for private sources |
Choose a source from Add skill, then follow its flow:
- Scan local: wait for candidates, filter with search, and select the skills to import. Rescan when the list needs refreshing.
- ZIP import: choose a skill archive and inspect the import result and errors; an arbitrary document archive is not a skill package.
- GitHub sync: enter
owner/repo, a branch or tag, and the skill subdirectory. A blank branch uses the default branch. Select Sync. - Review the same-name conflict choice after comparing sources and instructions. For local scanning, use Import selected.
- Review any script or high-impact capability confirmation before continuing. Inspect added, skipped, and failed entries afterwards; fix failures rather than importing the entire source again.
To write a skill from scratch, choose Create skill under Add skill. Fill in the ID, version, description, and SKILL.md body, then select required command execution, file writes, web reading, search, and allowed domains. Versions use three numeric segments such as 0.1.0; an existing skill’s ID is fixed during editing. Reopen the detail after saving to inspect the content. Declared capabilities do not replace runtime authorization.
See Import experience for the first-time migration overview.
Inspect content and dependencies
- Read the trigger, inputs, steps, and output.
- Inspect scripts, network access, write locations, and paid services.
- Prepare required plugins, MCP servers, command-line tools, and accounts.
- After import, check the source, content, and activation state in the list.
Instructions do not authorize actions, install dependencies, connect accounts, or expand tool permissions.
Verify readiness
Run a task on bounded material and check skill selection, available dependencies, and the expected output. Check both valid input and missing required input. Successful import does not prove workflow completion.
Inspect skill-use records and actual steps in the task. Mentioning a skill name in the final reply is not proof of use.
Update, disable, and restore
Preserve the original content and source before editing. Read SKILL.md in the skill detail. For a personal skill, edit its text, save, and reopen to verify; Cancel leaves the draft unsaved. For a GitHub source, use its sync flow and compare the branch, subdirectory, instructions, and scripts before importing changes. Review new data transfers, writes, or costs. Disable an unused installed entry from its list controls and verify again after reenabling it.
Repeat the original validation task after an update. If it fails, disable the entry, restore preserved content, and check dependencies. Do not assume automatic recovery without a recycle-bin entry; check the object and confirmation scope before deletion.
Troubleshoot
| Symptom | Check |
|---|---|
| Missing after import | Import result, source path, directory structure, and list refresh |
| Not selected by the task | Activation and whether instructions match the task |
| Selected but execution fails | Tools, accounts, scripts, dependencies, and inputs |
| Results change after an update | Compare content and dependencies; repeat the same task |
Put recurring project requirements in Project working rules, and account connections in Plugins and app connections.