Output Format Files
An output format file tells the agent how to structure its responses — table titles, insights, disclaimers, data context, and tone. It controls what every answer looks like, not what the answer says.
Frontmatter
---
type: behavior
kind: output_format
domain: "*" # or a specific domain name
---type
behavior
Identifies this as a behavior file
kind
output_format
Marks this as an output format file
domain
"*", "default", "marketing", etc.
Which domain(s) this applies to
Default Behavior
If no output format file is configured, Lynk applies its built-in output format — a sensible default shared across all Lynk customers. Adding an output format file overrides that default for your project. You are not starting from scratch — you are customizing on top of an existing baseline.
What Goes in an Output Format File
Tone — how the agent should sound. Professional? Direct? Casual? Warm?
Tables — whether to add titles, captions, and what context to include below a table.
Insights — how many key takeaways to surface after data, and how to format them.
Analysis — whether to show percentage of total alongside absolute numbers, and what metadata to always include (filters applied, date range, known caveats).
Examples — annotated sample responses make the format concrete and testable.
Custom Domains
A marketing domain output format file applies only to marketing-domain queries. Output format not defined in a custom domain falls back to the domain: "*" file.
Best Practices
Include annotated examples. The most effective way to communicate output format is to show it. An example response is worth ten rules — the agent can infer format from a well-crafted example more reliably than from abstract instructions.
Guide style, don't script every word. Define the structure (title, insights, data notes) and the tone (direct, warm, formal). Leave the agent room to write naturally. Over-specified files produce formulaic responses that feel robotic.
Define what metadata to always show. Users shouldn't have to ask "what date range is this?" or "are trials included?" Decide once — filters applied, date range, key caveats — and put it in the format file.
Keep tone instructions to one sentence per rule. "Lead with the answer" is actionable. "Always be helpful, concise, professional, and engaging while avoiding unnecessary preamble" is not.
Use domain-specific files for audiences that need different formats. An executive audience and an analyst audience want different levels of detail. Define a domain-specific output format file rather than trying to handle both in one.
Common Pitfalls
Over-specifying every response — very long, highly specific output format files produce rigid, formulaic responses. Guide style and structure — do not script every word.
Putting SQL rules here — SQL guidance belongs in task instructions. The agent does not apply output format rules when writing SQL.
Putting business definitions here — definitions belong in knowledge files. Output format files are purely about how responses are structured and presented.
When to Use This File
Create or update an output format file when a situation meets one of these conditions:
Agent responses are inconsistent in structure and you want a standard format
A domain has different audience expectations — technical vs. non-technical, formal vs. casual
Every response should include specific elements regardless of the question (titles, data notes, insights)
Examples:
"Answers should always end with a 'Data notes' section showing which filters were applied and what date range was used — users keep asking" → add a Data notes rule
"Every table response needs a short title and 2–3 bullet insights below it — right now it's inconsistent" → define table and insights structure
"The agent gives long, chatty responses — we want short and direct, lead with the answer" → set tone rules
"The marketing domain should feel warmer and more exploratory; the analytics team wants precision and no filler" → create a domain-specific output format file
"Responses should always show percentage of total alongside absolute numbers when showing a breakdown" → add an analysis rule
When NOT to Use This File
If you want to control what the agent knows (definitions, business rules, data caveats) → knowledge file
If you want to control when the agent asks questions before answering → clarification policy file
If you want SQL execution rules (filters, field choices, query patterns) → task instructions file
If you want to define business terms the agent should recognize → glossary file
Full Examples
Example 1 — Grove (B2B SaaS), analytics team
Structured and information-dense. The Grove analytics team values precision — they always want to understand the scope and assumptions behind an answer.
Example 2 — Bly (E-commerce), marketing and growth team
Conversational and action-oriented. The Bly growth team cares about what to do next, not just what the numbers say.
Example 3 — Arcadia (Mobile gaming), live ops team
Numbers-first and comparison-oriented. The Arcadia live ops team makes daily decisions based on metric deltas — they need absolute numbers and prior-period comparisons in every response.
Last updated