Naming Conventions

Why this exists

Consistent naming makes it immediately clear what a prompt does, who calls it, and where it belongs — without opening the file. It also makes bulk operations (search, refactor, agent edits) predictable and safe.

Rules

Prompt files follow the pattern:

{prefix}_{domain}_{action}.yaml

All parts are lowercase, separated by underscores. No camelCase, no dashes.

Prefixes

Prefix	Meaning
`flow_`	Sequential step in a Dify workflow pipeline (LLM node)
`tool_`	Callable utility invoked on demand inside a workflow (e.g., image generation, web search)
`agent_`	Prompt defining agent behavior or planning logic
`fraya_`	Prompt specific to Fraya's core identity and behavior

Injections do not use prefixes — their type is signaled by location (/prompts/injections/).

Note: flow_ vs agent_ — flow_ is the current default for workflow steps. agent_ may replace it in the future for steps that involve reasoning or planning. Discuss before switching.

Domains

Domains reflect the subject area of the prompt within Fraya's system.

Domain	Scope
`audience`	Learner profiling and audience analysis
`depository`	Knowledge Depository — building and trimming the knowledge foundation
`course_structure`	Outline, module structure, and course-level planning
`asset`	Per-course assets generated once after the outline: characters, localization, publishing blocks
`section_content`	Content generation for course sections
`artifact`	Interactive mini-game generation pipeline
`avatar`	Trainer personas
`video`	Synthesia templates and scripts
`image_translation`	Image localization workflows

Actions

The action describes what the prompt does, as a verb or verb phrase.

Examples: generate, check_quality, plan, analyze, render, select, define_profile

Examples

flow_audience_define_profile.yaml              ✓
flow_audience_analyze.yaml                     ✓
flow_depository_plan.yaml                      ✓
flow_depository_fill.yaml                      ✓
tool_image_translation_generate.yaml           ✓
agent_plan_work.yaml                           ✓
fraya_core_behavior.yaml                       ✓

imageTranslation.yaml                          ✗  no prefix, camelCase
generate_prompt.yaml                           ✗  no prefix, no domain
prompt_1.yaml                                  ✗  meaningless name

Edge cases

Prompt serves two domains Pick the primary domain — the one that best describes the prompt's main responsibility. If truly ambiguous, discuss before naming.

No clear action If the prompt is a configuration or a definition rather than an action, use a noun: flow_artifact_framework.yaml, flow_course_structure_constraints.yaml

Injection naming Injections are named after what they enforce, not what they do: avoid_ai_language.yaml, html_instructions.yaml, language_rules.yaml

Course content naming

Rules for naming modules, topics, and sections inside a generated course. This is separate from prompt file naming and enforced at the outline design stage via the naming_instructions injection.

General rules

Use clear, simple language. No jargon, buzzwords, or overly long titles.
Consistent tone: professional, educational, encouraging.
Sentence case throughout: capitalize only the first word, proper nouns, and abbreviations.
Titles must connect to the course title and reflect its main idea.
Never use the word "journey".
Do not use titles in the form "[name]'s aha moment" — rephrase to something equivalent.

By level

Level	Length	Form	Notes
Module	2–4 words	Noun phrase	Express the main theme or skill area (e.g., Recognizing strengths, Smart AI workflows)
Topic	max 6–8 words	Action-oriented where possible	Natural step in the learning path; avoid "Module X summary" — reflect the theme instead
Section	—	—	Must differ from the section type name; no numbers, no underscores; avoid "Module X challenge"