Pro — Agent Skills are a Pro feature. Upgrade in the app.

Agent Skills are local SKILL.md instruction packs that Termorq can include in model requests. This is separate from the / prompt palette: those entries are Termorq prompt recipes; Agent Skills are portable instructions discovered from folders on your account.

Choose sources and defaults

Open Settings → AI → Skills. Skill discovery is on by default and reads these folders under the selected home folder (blank uses your operating-system account home). Select Refresh after adding or changing skill files to rescan the enabled folders. Providers use distinct coloured pills; each skill shows a short description you can expand, and Common defaults starts collapsed:

SourceFolder
Shared Agent Skills.agents/skills
Claude.claude/skills
Copilot.copilot/skills
Codex.codex/skills
Grok.grok/skills
Cursor.cursor/skills

Switch off a source to stop discovering skills there. Common defaults are included in every enabled AI area. Each area can use Auto select, Use selected, or Off for Ask and Agent, workflow AI blocks, pipeline explanations, and other AI calls. Auto sends the request plus skill names and descriptions to your configured model for a relevance choice. It loads only the selected Markdown after validating the returned IDs. This adds a small model request and can add latency or provider cost.

Workflow nodes

An AI block inherits the workflow policy from Settings. In the block properties, Skills for this node can inherit that policy, use explicit skill IDs, or turn skills off for that node. In YAML, the same setting looks like this:

blocks:
  - id: review
    kind: ai
    connection: Local model
    skills:
      mode: only_selected
      include:
        - claude:dotnet/review
    prompt: Review prior output.

Use mode: inherit to keep the area defaults or mode: none to suppress skills for one AI node. Each completed block record shows the skill name, source, selection reason, and a short SHA-256 digest. The full skill body is not copied into the run record.

What a skill can contain

Each package has a SKILL.md file with YAML frontmatter and Markdown instructions:

---
name: dotnet review
description: >
  Review .NET changes for
  correctness, security, and tests.
---

Check error paths and cancellation.
Suggest focused tests for changed
behavior.

Termorq reads bounded Markdown and same-package .md references only. It does not execute scripts, import tool permissions, or follow links outside the package. A skill marked disable-model-invocation: true, or a Codex skill whose agents/openai.yaml disables implicit invocation, is excluded from Auto but can still be selected explicitly. At most four skills and 8,000 estimated tokens are sent with one request.

Skill instructions are user-provided model context. Review a pack before selecting it, and remember that its text is sent to the AI connection you chose. API keys and known secrets are scrubbed from the task text before skill selection; skill folders and their contents stay on this computer.

When an AI request includes skills, a compact Skills used disclosure appears beside the call. It shows skill names and providers on one quiet line; expand it to see why each was selected, its skill ID, and a short content hash. Workflow runs keep the same details on the AI block, so you can inspect what was included after the run finishes. If automatic selection fails, the disclosure explains why the AI call continued without optional skills.