Skilder Logo
Skills & Hats

Assets & References

Attach instruction documents, binary files, and downloadable assets to your Skilder skills.

Skills aren't just tools and code. You can attach references (instruction documents) and assets (binary files) that give agents the context they need to do their job well.


References

References are markdown documents that provide instructions, SOPs, domain knowledge, or any structured guidance for the agent. Think of them as the "how-to guide" that comes with the tools.

When to Use References

  • Step-by-step workflows the agent should follow
  • Domain-specific terminology or rules
  • Output format specifications
  • Constraints and guardrails

Creating a Reference

  1. Open a skill in the Skill Editor.
  2. In the file tree, right-click References > Add Reference.
  3. Give it a name (e.g., "Scouting Process") and a description.
  4. Write your content in markdown.

Tips for Good References

  • Be specific. "Summarize in 3 bullet points" beats "keep it short."
  • Structure with headings so agents can scan for relevant sections.
  • Include examples of expected output when possible.
  • Use constraints: "Never include PII" or "Always cite the source URL."

Assets

Assets are binary files — images, PDFs, spreadsheets, or any document up to 10 MB — that you attach to a skill. Agents can download and use these files during execution.

When to Use Assets

  • Template files agents should fill in (Excel, CSV)
  • Reference images or diagrams
  • PDF documentation that agents need to read
  • Sample data for testing or demonstration

Creating an Asset

  1. Open a skill in the Skill Editor.
  2. In the file tree, right-click Assets > Add Asset.
  3. Give it a name and description.
  4. Upload the file (max 10 MB).

Assets are stored as base64-encoded data and are available to any agent using the skill.


The @skilder-asset:/ Path Scheme

When an agent downloads an asset during execution, Skilder returns a special path:

@skilder-asset:/<resourceId>/<filename>

This path can be passed to other tools that accept file paths. The runtime automatically resolves it to a local cached file.

How It Works

  1. Agent calls the download_asset tool with a skill path (e.g., /My Skill/template.xlsx).
  2. Skilder fetches the asset from the backend and caches it locally.
  3. The tool returns a @skilder-asset:/ path.
  4. Other tools can use this path — the runtime resolves it to the actual file on disk.

Example

Agent: "I need the report template"
→ calls download_asset with path "/Sales Reports/quarterly-template.xlsx"
→ gets back @skilder-asset:/abc123/quarterly-template.xlsx
→ passes this path to another tool that fills in the spreadsheet

Cache Behavior

  • Assets are cached locally for 1 hour (configurable via SKILDER_ASSET_CACHE_MAX_AGE_MS).
  • Stale assets are garbage-collected automatically.
  • Cache location defaults to /tmp/skilder/assets/.

Skill Components Summary

ComponentFormatPurposeLimit
ReferencesMarkdownInstructions, SOPs, domain knowledgeNo size limit
ScriptsPython, Bash, JS, TSExecutable code with SDK access30s timeout
AssetsAny binary fileTemplates, images, PDFs, data files10 MB per file

All three are managed in the Skill Editor file tree.


Next Steps