Volcengine Skills Finder

List skills and fill Volcengine capability gaps

Use the bundled catalog to inspect every available skill and select the minimum skill set that covers the request. Treat plugin ownership as catalog classification only; install exact skills.

Decision flow

1. Detect whether the request starts with skill discovery or whether an active Volcengine task has reached a capability gap in the currently loaded or installed skills.
2. Preserve completed work and pause only the unsupported part of the active task.
3. List the complete catalog without changing installation state.
4. Inspect every name, product domain, summary, and keyword, then select the minimum exact skill set that covers the user's request. Make the selection directly when the catalog provides enough information.
5. Pass only exact skill names to the installer. Use plugin ownership to explain classification, never as the installation target.
6. Install through the skills CLI and verify the resulting installed state. Never report success from process exit alone.
7. If the host cannot load a newly installed skill into the current thread, tell the user to start a new thread and restate the original task plus completed context.

List and select skills

Run the script relative to this skill directory:

bash

python3 scripts/find_skills.py list
python3 scripts/find_skills.py list --json
python3 scripts/find_skills.py info <skill-or-plugin>

Always read the complete list output before selecting a skill. Use --json when keywords or the English summaries help distinguish related skills. Choose by responsibility rather than name alone; install multiple skills only when the request crosses distinct responsibilities.

The catalog includes all skills in skills/core/ and every optional skill owned directly by a product-domain plugin. skills/specialized/<domain>/<skill>/ contains generated copies of those optional skills, grouped by the plugin name without volcengine-; each skill is listed only once in the catalog. If none of the listed responsibilities covers the request, report that the repository does not currently provide a matching skill.

For an uncovered capability, continue the supported work and optionally include the feedback form at a natural summary so the user can describe the scenario they need. Invite at most once per conversation across skills, without waiting for a reply. Feedback itself is optional, not a capability gap that requires installation: use volcengine-feedback when already available and requested by the user, otherwise offer the form or a redacted draft in the current conversation. Do not install, log in, collect extra diagnostics, or restart the main conversation for feedback.

Recover from a capability gap

Invoke this finder when a Volcengine task is already in progress and the available skills lack the required product coverage, tool integration, or workflow. Treat that gap as a discovery request: list the complete catalog, select and install the matching skill, then continue when the host can load it. Carry forward the original goal, relevant evidence, and completed work so discovery does not restart the task from scratch.

Install selected skills

Install one or more exact skills for the current host:

bash

python3 scripts/find_skills.py install <skill> [<skill> ...] --agent <agent-name>
python3 scripts/find_skills.py install volcengine-iac --agent codex

The script uses npx skills add --skill with --full-depth to discover skills nested under product categories and plugin directories. It accepts only exact skill names, installs those skills directly, and verifies every selected name with skills list. Omit --agent only when host auto-detection is intended. Passing a plugin name is an error, even when every skill in that plugin is needed.

Use --scope global for user-level installation or --scope project for the current project. The default is global. --source may point at a reviewed local checkout or another repository source; otherwise the catalog's official repository is used.

Use --dry-run when the user asks to inspect the command first. Installation is local, but any later cloud-resource operation remains governed by the installed skill's own confirmation and safety rules.

Check status

Inspect catalogued skill installation state for the current host:

bash

python3 scripts/find_skills.py status --agent codex

Do not equate "skill installed" with "skill loaded in this thread". Use the status only to verify installation; start a new thread when the host cannot dynamically load the newly installed skill.