Method 1: Write instructions manually
Go to Skills in the sidebar and click Add Skill, then select Write skill instructions. Fill in the following fields:The description field
The description is critical. The AI reads it on every message to decide whether this Skill is relevant. A vague description causes the Skill to activate at the wrong time, or not at all.
Write the description as if explaining to a colleague when they should reach for this Skill. Include concrete trigger phrases.
Method 2: Create with chat
Go to Skills, click Add Skill, then select Create with chat. Describe what you want in plain language and Langdock generates the name, description, and instructions for you. This is the fastest starting point if you are not sure how to phrase the instructions. You can review and edit everything the AI generates before saving.Method 3: Upload a Markdown file or ZIP bundle
Upload a Markdown file to create a Skill without supporting files. For Skills that include supporting files, upload a.zip archive with a SKILL.md file.
Go to Skills, click Add Skill, then select Upload a skill and drag in your file.
Markdown file format
The Markdown file must contain a YAML frontmatter block at the top:
The instructions body (the content after the frontmatter block) is required.
ZIP bundle
A ZIP archive can contain aSKILL.md file plus supporting files such as scripts, references, or assets.
Structure: The SKILL.md must be at the root of the archive or inside a single top-level directory.
Allowed file types: .md, .mdx, .txt, .py, .sh, .js, .ts, .json, .yaml, .yml, .toml, .csv, .xml, .xsd, .html, .css, .svg, .pdf, .docx, .xlsx, .pptx, .png, .jpg, .jpeg, .gif, .webp
Size limits:
Edit an existing Skill
Go to Skills and select a Skill to open its detail view. If you can edit the Skill, you can update:- Name and icon in the header.
- Description and instructions by opening
SKILL.mdand clicking Edit skill. - Supporting files by adding new files and folders or uploading files. You can edit, rename, move, or delete existing supporting files.
- Integrations by clicking Edit integrations.
SKILL.md remains the main instructions and metadata file and cannot be renamed, moved, or deleted. Other Markdown files can use different names.
Writing good instructions
The quality of a Skill depends almost entirely on the quality of its instructions.- Be specific about the output. Do not say “summarize things.” Say “write a three-section summary with: (1) key decisions made, (2) open questions, (3) next steps. Each section should be 3 to 5 bullet points.”
- Write the description as if explaining to a new colleague. Vague descriptions cause the AI to activate the Skill at the wrong time, or not at all.
- Test before sharing. Install the Skill, start a new chat, and see if the AI behaves as expected. Adjust based on what you observe.
- Keep instructions focused. A Skill that does one thing well is more reliable than one that tries to cover many cases.
- Include concrete trigger phrases in the description. Examples: “when asked for updates,” “when drafting a reply to a customer,” “when summarizing a meeting.”
Limitations
What Skills cannot do:
- Skills do not have memory across sessions. Instructions are re-read fresh at the start of each conversation. For persistent memory, use Chat Memory.
- Skills cannot access live data on their own. Unless linked to an integration, a Skill only provides instructions and does not fetch real-time information.
- No version history. When you update a Skill’s instructions, the previous version is not saved.
FAQ
When should I create a Skill?
When should I create a Skill?
Create a Skill when you have reusable instructions that should guide future chats. Skills are best for repeatable work patterns, writing rules, domain context, or internal processes that are too small to require a full agent.
What should I check if a new Skill does not work well?
What should I check if a new Skill does not work well?
Check whether the instructions are specific, scoped, and testable. Avoid mixing too many unrelated goals in one skill, and test it with realistic prompts before sharing it widely.