Skip to main content
Skills are markdown files that give the Agent deep, task-specific knowledge for a particular type of work. Define one for short story creation, one for analysis the book, one for screenplay formatting — and the Agent loads only the skill that’s relevant to what you are doing right now.

​
How skills work

Skills use progressive loading to keep context lean. When you open a project, the Agent sees a list of available skills with their names and descriptions — a few dozen words each. That is all it loads at startup. When you ask for something that matches a skill’s description, the Agent activates that skill and loads its full instructions. A skill for short story creation stays dormant while you are working on a query letter. A skill for horror fiction does not consume context when you are outlining a thriller.

​
Where skills live

Every skill is a directory containing a SKILL.md file. The directory name becomes the skill name. Skills can also include supporting files in three optional subdirectories.
Supporting files are loaded only when the instructions reference them — they do not consume context at startup. A template stays dormant until the Agent reaches the step that calls for it. A reference document is read only when the Agent needs to look something up. Global skills are stored in the system-level skills directory and are available across all your projects. Use them for general-purpose capabilities you apply consistently — a skill for generating opening hooks, or one that guides the Agent through your standard manuscript review process. Workspace skills live in .soloent/skills/ at your project root. Use them for capabilities tied to a specific project or genre — a skill built around the particular rules of your romance series, or one calibrated to the explicit level and narrative register of your current story.
When a global and workspace skill share the same name, the workspace skill takes precedence.

​
Toggling skills

Every skill has a toggle to enable or disable it without deleting the file. Disable a skill when you do not need it for the current task — this keeps it out of context and avoids unintended activation.

​
Creating a skill

The SKILL.md file has two parts: a frontmatter block (metadata) and the instruction body.
Two required frontmatter fields:
  • name must exactly match the directory name
  • description describes when to trigger this skill, not what it does (under 1,024 characters)
SKILL.md file open in the editor showing the frontmatter fields name and description, with the Skills panel on the right displaying the english-short-story skill listed under Workspace Skills
1

Open the Skills menu

Click the fourth icon from the left in the Agent panel，and switch the Skills manager.
2

Create a new skill

Click “New skill…” and enter a name (e.g., short-story) and click on ”+” . The Agent creates a skill directory with a template SKILL.md file.
3

Write your skill instructions

Edit the SKILL.md file:
  • Update the description to specify when this skill should trigger
  • Add your instructions, checklists, and workflow steps in the body
  • Optionally add supporting files in templates/ or docs/ subdirectories
This is the official english-short-story skill file — a domain expert type skill with a complete 6-step workflow, checklists, and stop conditions. Copy and paste it directly.

​
Three common skill types

​
Type 1: Checklist type

Best for verifying multiple criteria at a fixed point — so nothing gets missed. For example, a pre-submission check after each chapter is done:

​
Type 2: Workflow type

Best for operations with a fixed sequence of steps and a confirmation step before execution.

​
Type 3: Domain expert type

Best for situations where you want the Agent to work from expert experience rather than guessing. For example, a diagnosis skill for when the writing gets stuck:

​
Full example: Book summary

This is a domain expert skill. It defines a content-retrieval path (a script extracts chapters on demand), specialized knowledge (chapter recognition patterns in docs/), and a fixed output format (a note template in templates/) — the three elements that characterize a domain expert skill. This skill extracts structured reading notes from a local book file — chapter summaries, key insights, and quotations — and saves them as formatted Markdown.

​
Directory structure

​
Why this skill needs a script

Books are long. Loading an entire novel into context to extract its chapter structure would be wasteful and slow. The preprocessing script solves this:
  • info command — reads the first 50 lines to identify the title, encoding, and total line count
  • chapters command — scans the file for chapter headings and returns their line numbers, without reading the full text
  • extract command — reads only the lines for a specific chapter, so the Agent processes one chapter at a time
The Agent calls the script at each stage of the workflow. Only the script’s output enters context — not the book itself.

​
Why this skill needs a docs file

Chapter heading formats vary widely across Chinese and English books. Rather than embedding all the recognition patterns in SKILL.md, the skill stores them in docs/chapter-patterns.md. The Agent reads this reference file when it needs to identify chapter boundaries — and ignores it otherwise. Keeping reference material in docs/ has two benefits: it keeps SKILL.md focused on workflow logic, and it lets you update the patterns without touching the main instructions.

​
Why this skill needs a template

The output format is fixed: every set of reading notes follows the same structure — book overview, chapter summaries with key points and quotations, book-wide highlights, and a personal reflection section. Rather than re-specifying this format in every prompt, the skill stores it in templates/note-template.md. When Step 5 (final assembly) runs, the Agent loads the template and fills in the placeholders. The format is consistent every time, and you only need to maintain it in one place.

​
How the SKILL.md references supporting files

The instructions in SKILL.md point to each supporting file explicitly:

​
Writing effective skills

​
Write “when to use me”, not “what I do”

These sound similar, but they have a significant impact on trigger accuracy.
Descriptions stay in context at all times — every enabled skill is continuously consuming your token budget. Keep them short, keep them specific, and state the trigger scenario directly:

​
Complete steps, inputs, outputs, and stop conditions

A skill instruction that only writes the beginning of a process — with no clear completion signal — leaves the Agent to decide when to stop. That is rarely where you want it to stop. Every step should have:
  • Input: what context or files this step needs
  • Output: what this step produces
  • Stop condition: when to wait for confirmation, when to continue

​
Keep the body lean — move bulk material to supporting files

Keep SKILL.md under 5,000 words. If you need large reference material — style guides, recognition patterns, character profiles — put them in the docs/ subdirectory and reference them by filename. The Agent loads them only when needed, keeping them out of context otherwise.

​
Decide auto-trigger based on call frequency

​
Common pitfalls

​
Troubleshooting

Most likely cause: The description isn’t specific enough, or the skill is toggled off.
  • Check that the skill is toggled on in the Skills panel
  • Rewrite the description to include the exact phrases and task types you use — “helps with writing” rarely triggers correctly; name the workflow, the output, and sample invocation phrases
  • Make sure the skill directory contains a valid SKILL.md with both name and description frontmatter fields
  • Verify that name in the frontmatter exactly matches the directory name
The description is too broad — it matches requests you didn’t intend it for.
  • Narrow the description to describe only the specific tasks this skill handles
  • Toggle the skill off when you’re working outside its domain
  • If you have multiple skills with overlapping descriptions, make each one more specific about when it should and shouldn’t activate
Check for these common issues:
  • The instructions may exceed the context limit — keep SKILL.md under 5,000 words and move reference material into docs/ subdirectories
  • A conflicting global or project rule may be overriding part of the skill’s behavior — check your active rules for contradictions
  • If the skill uses checklists, ensure each checklist item has an explicit “wait for confirmation” gate; without it the Agent may advance automatically
  • Supporting files referenced in the instructions must exist at the specified paths — a missing file causes that step to be skipped silently
The Agent only loads supporting files when the instructions explicitly reference them.
  • Check that the file path in the instruction matches the actual location (e.g., templates/chapter-outline.md, not just chapter-outline.md)
  • Verify the file exists inside the skill directory under the correct subdirectory (templates/, docs/, or scripts/)
  • If the file is never referenced by name in SKILL.md, the Agent has no reason to load it
Workspace skills always take precedence over global skills of the same name.
  • This is expected behavior — if a workspace skill is active, the global skill with the same name is ignored entirely
  • If you want the global skill to apply, rename or remove the workspace skill
  • If both should coexist, give them distinct names and update their descriptions accordingly

​
More advanced tips

Workflows

Automate repeatable multi-step processes with workflow files

Rules

Capture your writing preferences so the Agent always writes in your voice

Rules vs. Workflows vs. Skills

Three ways to instruct your Agent — learn the differences and when to use each