docs
v0.7.8

Priorities & Kinds

Priorities and kinds control how the agent approaches different types of work. Each has a methodology prompt, reasoning effort level, and optional temperature override.

Priorities

Priorities set the urgency and depth of agent work. Configure them in Admin → Prompts & Model → Priorities.

Default priorities

ID Label Reasoning Effort Behavior
urgent Urgent Low Fastest correct fix. Minimize exploration. Skip optional improvements.
high High Medium Efficient and thorough. Clean solution with proper error handling.
medium Medium Medium Balanced thoroughness and efficiency.
low Low High Take time to do it well. Explore broadly. Consider refactoring. Write thorough tests.

Each priority has a prompt that is injected into the agent's system prompt when a ticket uses that priority.

Reasoning effort

Reasoning effort maps directly to the AI model's reasoning parameter:

  • Low — Quick, less deliberation. Agent responds faster but may miss edge cases.
  • Medium — Balanced. Good default.
  • High — Deep reasoning. Agent takes longer but handles complex problems better.

Match reasoning effort to the priority. Urgent tickets should use low effort (speed matters). Low-priority tickets can use high effort (quality matters).

Kinds

Kinds define the type of work and the methodology the agent follows. Configure them in Admin → Prompts & Model → Ticket Kinds.

Default kinds

ID Label Reasoning Effort Methodology
bug Bug High Reproduce → Diagnose → Fix → Test → Verify. Minimal changes, correctness over elegance.
feature Feature Design → Plan → Implement → Test → Document. Build incrementally.
task Task Understand → Plan → Implement → Test → Summarize.
chore Chore Assess → Plan → Execute → Verify. Systematic, atomic changes.

Kind-specific overrides

Each kind can optionally override:

  • Reasoning effort — Overrides the priority's reasoning effort. For example, bug defaults to high regardless of priority, because bug diagnosis benefits from deep reasoning.
  • Temperature — Overrides the global temperature. Useful if a particular kind of work benefits from more or less randomness.

If a kind doesn't set an override, it inherits from the ticket's priority.

Adding custom priorities and kinds

Click Add in the priorities or kinds section. Each entry needs:

  • ID — A lowercase identifier (e.g., security_audit). Cannot be changed after creation.
  • Label — Display name in the UI.
  • Prompt — Methodology instructions injected into the agent's system prompt.
  • Reasoning effort — Low, medium, or high.
  • Temperature (kinds only) — Optional override.

AI-generated descriptions

When Codality starts with an OpenAI API key configured, it automatically generates concise AI descriptions for each priority and kind. These descriptions help the AI understand the intent behind each option when making decisions.

How priorities and kinds interact

When the agent starts on a ticket with priority high and kind bug:

  1. System prompt includes the bug methodology prompt
  2. System prompt includes the high priority prompt
  3. Reasoning effort = high (bug kind overrides the priority's medium)
  4. Temperature = global default (bug kind doesn't set a temperature override)

The kind prompt tells the agent what process to follow. The priority prompt tells it how much effort to spend.