A good AI-assisted run starts with a well-prepared workspace. This guide covers what to put where before FilepadAI or an external agent uses it.
Start with sources/
sources/ is grounding material: documents, data, and references FilepadAI or external agents may reason from. Before your first run:
- Upload the documents the work should be grounded in
- Use descriptive filenames (date + topic is a reliable default)
- Remove obvious duplicates
- Prefer final versions over drafts or long email threads
You can add more sources between runs. FilepadAI and external Agent Access read the current file tree.
Add external-agent operating notes
If you plan to connect an external agent, create an optional note under agents/, such as research-agent.md. Cover:
- What the external agent is: name, model, purpose
- What it should do and how it should behave
- What it is explicitly not supposed to do
- The Agent Access scope it has been granted
Be specific. Vague instructions produce vague behavior.
Build skills/ incrementally
skills/ holds reusable prompt modules. Start with the most common tasks:
skills/
summarize.md
extract-entities.md
format-report.md
Each skill is a Markdown file. Write it the way you would write a good prompt, then give it a clear filename. External agents discover skills via the /prompts endpoint and can call them by name.
Don't over-engineer skills before you've run the agent. Write one, observe the output, refine.
Curate memory/
memory/ is for durable workspace context that should not require restating in every prompt:
- Project-specific terminology or naming conventions
- Decisions that have been made and should not be revisited
- Standing preferences (
output_format: markdown,language: formal) - Facts FilepadAI or external agents should treat as ground truth
Keep memory/ small and explicit. Long, vague memory files are worse than no memory.
Leave artifacts/ and automations/ empty at first
artifacts/ fills up as work happens. You do not need to pre-populate it.
automations/ is for reviewed automation definitions. Leave it empty until you have repeating work you want to codify.
Check the environment endpoint before the first external-agent run
After setup, verify what the external agent will see:
GET /agent-api/v1/workspaces/:workspaceId/environment
This returns the canonical folder structure and what is visible. If something is missing or misplaced, fix it before the external agent connects.
Next: Connecting an agent.