SharePoint File Creation

Generate files in a Copilot Studio session and upload them to SharePoint without passing raw file bytes through the model. Use this workflow instead of reading binary file content into the model, converting files to base64, or passing raw bytes through connector arguments.

This skill gives an agent a fixed, safety-focused workflow:

1. Generate the complete requested file in Copilot Studio’s /app/created/ directory.
2. Measure the final file with the filesystem API and stop before upload if it is empty or 4,000,000 bytes or larger.
3. Pass the file path directly to the configured SharePoint Create file action so the connector resolves the file server-side.
4. Report the SharePoint destination, measured pre-upload size, and upload result.

Why the path and size rules matter

The workflow is based on connector testing in Copilot Studio:

• /app/created/<filename> is the confirmed path-reference mechanism for the Create file action. Passing a path outside that directory can upload the literal path text instead of the intended file bytes.
• Uploads begin failing around the tool layer’s 4 MiB inline-upload boundary. The skill uses a conservative 4,000,000-byte gate because the complete range immediately below 4 MiB has not been verified.

These are empirical, environment-specific findings rather than documented SharePoint limits. SharePoint itself supports larger files; this skill addresses the current inline connector/tool behavior available to the agent.

Before adding the skill

Configure a SharePoint Create file action for the agent. Fix the site and folder in the action when every upload should use the same destination, or leave them available as inputs when the destination must vary.

Review Section 0 of SKILL.md and replace or extend the generic content-sourcing rules when the agent must use a particular template, knowledge source, report structure, schema, or style guide. The upload and size-gating sections should remain unchanged.

Boundaries

• It does not automatically truncate or split oversized output.
• It stops on permission, DLP, authentication, ambiguous-result, size, or integrity errors rather than silently retrying.
• The raw sandbox path stays internal during normal use.