Skills & agents

Workspace context

A single document every skill reads first. Your product, audience, brand voice, words to never say. Set it once and every skill — built-in or custom — stays on-brand without you re-pasting the context into every prompt.

#Why you want this

Without workspace context, every skill run starts cold. The LLM has no idea what you make, who you sell to, what tone you use. You end up either:

  • Pasting brand info into every prompt (tedious, easy to forget), or
  • Accepting generic, off-brand output ("Welcome to your dashboard!" when your brand is sharp, terse, cynical).

Workspace context fixes this. It's a structured document loaded into the prompt for every skill run, automatically.

#What goes in it

The context has 8 fields, each optional. Fill in what's relevant.

FieldTypeDescription
productstringWhat you make. One sentence.
audiencestringWho buys it / uses it.
voicestringHow you sound. Adjectives + 2–3 example sentences.
wordsToAvoidstringA comma-separated list. "synergy, leverage, unlock value".
wordsToPreferstringWords your brand uses where others wouldn't.
brandColorsstringHex codes or descriptive names. Used by visual augmenters.
positioningstringHow you compare to competitors. The pitch.
noGoTopicsstringThings you never want surfaced. Compliance limits, sensitive areas.

#Two layers: team + personal override

Workspace context has two layers that merge at run time:

  • Team layer — set by anyone with access to the workspace. The shared baseline.
  • Personal override — your individual tweaks. Useful for testing a different voice or adding personal-only no-go topics. Only visible to you.

When a skill runs, your personal override is layered on top of the team layer. Empty override fields fall through to the team defaults.

#Editing the context

Two entry points:

  • Inline drawer in chat — click the workspace pill in the chat header. A side drawer opens with the 8 fields and a sync-status indicator.
  • Skills hub — at /skills, scroll to the "Workspace context" card.

#How skills consume the context

Every skill's system prompt has the workspace context injected near the top, before user input:

Skill prompt assembly
<system prompt for the skill>

# Workspace context
- Product: <product>
- Audience: <audience>
- Voice: <voice>
- Words to avoid: <wordsToAvoid>
- ...

# User message
<your prompt>

This means even a generic skill like /copywriting will rewrite copy in your brand voice, automatically. Same for augmenters: /seo picks brand-aligned meta descriptions,/marketing writes lead-form copy that sounds like you.

#What good context looks like

  • Be specific — "confident, plainspoken, never corporate" beats "professional".
  • Show, don't tell — voice is best with example sentences, not just adjectives.
  • Update as you learn — when a skill output misses your voice, identify the pattern and add a "words to avoid" entry.
  • Don't dump — keep it scannable. The whole document is loaded into every skill prompt, so brevity matters for both quality and cost.

#Where to go next

PromptFloe docs · last updated Jun 2026Report a doc issue