For AI agents: the complete documentation index is available at /en/llms.txt, the full documentation bundle is available at /en/llms-full.txt, and this page is available as Markdown at /en/guide/fun/menu/prompt.md.
  • English
  • /prompt

    Use /prompt to inspect, preview, and adjust the prompt assembly result for the next new session.

    This is a relatively core feature of Hunea. It's already usable and will continue to be polished.

    Open it when you need to confirm which sources the system prompt is assembled from, temporarily toggle instructions/skills/tool guides on/off, add custom prompts, or preview the final complete system prompt before modifying. After typing /prompt and confirming, it opens a full-screen panel titled Prompt Assembly.

    The menu description is Inspect prompt assembly for the next new session. Most changes do not rewrite the transcript for the current already-started session; they target the "next new session". If the current session is still empty, closing and persisting may apply directly to the current empty session (see "What happens after closing" below).

    Panel overview

    /prompt is a two-column manager, not a single list:

    1. Title bar: left is Prompt Assembly, right has four candidate tabs
    2. Left column Active: sources already included in "next new session assembly"
    3. Right column (candidates / inventory): candidate collections not fully assembled, or toggleable
    4. Bottom pagination line: shows position on the currently focused side, e.g. Active (2 of 8)
    5. Bottom operation hint: changes with the currently selected item; ? more is often on the far right

    It roughly looks like this:

    prompt

    Left column: Active

    The header columns are roughly:

    Sel · Ord · Source · Type · Scope

    ColumnMeaning
    SelWhether enabled (checked state)
    OrdOrder in the assembly result
    SourceTitle / name; may have missing, shadowed, +N shadowed tags when needed
    TypeSource type, see table below
    ScopeScope, commonly project / global; built-ins have their own

    Type roughly corresponds to:

    DisplayWhat it is
    systemCore system prompt (built-in default, can be overridden by global / project override)
    instructionsInstruction files, like global AGENTS.md, project AGENTS.md / CLAUDE.md
    customCustom extra prompt
    discoverySkill discovery fragment (generated)
    skillLong-lived injected skill
    toolsTool usage guidelines (generated)
    dynamicDynamic environment information (base / changes)

    The currently focused row has a color swatch mark on the left. If an item "shadows" other sources with the same name/key, it may show +N shadowed; press Ctrl + E to expand/collapse these shadowed details.

    Right column: four tabs

    Top-right tabs (defaults to Skill):

    TabContent
    SkillDiscovered long-lived skill candidates
    Custom PromptsCustom prompt candidates; can create new
    ToolsTool candidates / guideline fragments; per-tool enable/disable (see below)
    DynamicDynamic environment sources; can toggle Base / Change separately

    Headers differ slightly between tabs:

    • Skill: Sel · Ord · Name · Scope
    • Tools: On · Guide · Ord · Name · Scope (On/Guide are dual-column switches, see below)
    • Custom Prompts: Sel · Name · Scope
    • Dynamic: Base · Change · Source · Scope

    Tools tab: On / Guide dual columns

    Each row on the Tools tab has two checkbox columns:

    ColumnMeaning
    OnWhether the tool itself is enabled. When off, the tool won't appear in the tool list sent to the model, so the model can't call it
    GuideWhether the tool's usage guidelines are injected into the system prompt

    How to operate: / moves between the On / Guide columns (the active column is highlighted), and x toggles the switch in the current column. Note: when a tool's On is off, its Guide column becomes non-operable (dimmed) and x is a no-op there; tools without guidelines also have a no-op Guide column. Disabled tools are filtered from both the model request and the session tool registry, and the state is persisted (tool_enablement). See Per-tool enable / disable.

    It looks roughly like this:

    tool-off-on

    How to operate

    Focus and navigation

    • / , or h / l: switch focus between left and right columns
      (on the Dynamic tab, left/right may also move between Base / Change columns)
    • / , or j / k: move selection up/down
    • PageUp / PageDown: page through
    • Home / End: jump to start/end of current list
    • Tab / Shift + Tab: when focus is on right column, cycle through top-right tabs
    • ?: toggle more shortcut help (the More overlay)
    • Esc: close the /prompt panel (triggers the commit/persist flow, see later)

    You can also use the mouse:

    • Click a top-right tab: switch to that candidate page
    • Click a left/right list row: switch focus and select that row

    Preview

    • Space: preview the content of the currently selected item (single source / skill / tool / dynamic, etc.)
    • p: preview the fully assembled system prompt (titled Assembled prompt)

    In preview mode:

    • Esc / Space / p: go back to the main panel
    • / / / , and h / j / k / l: page through longer content

    It roughly looks like this:

    prompt-preview

    prompt-assembled

    Editing, toggling, reordering

    The bottom hint changes with the selection. Common actions:

    ShortcutAction
    eOpen the currently editable item in external editor (footer may also say e/ctrl+g edit)
    aCreate a new custom prompt (first select Project / Global scope)
    dDelete the currently deletable item
    xToggle enabled/disabled (footer often shows x disable)
    J / K (Shift):Adjust order on the Active side (move down / up)
    rDepends on selection: restore core system override, or reset discovered skill order
    Ctrl + EExpand/collapse shadowed details

    Not every row supports all actions. For example:

    • Core system, instruction files, generated discovery / tools / some dynamic sources usually cannot be directly deleted with d like custom prompts
    • Only editable sources show e
    • Only on the Active side, when reordering is allowed, will you see J/K reorder

    Creating a new custom prompt

    Press a when in the Custom Prompts context on the right:

    1. First pops up scope selection: [Project] Global (currently project, global optional) or Project [Global] (currently global, project optional)
    2. / (or h / l) switch scope
    3. Enter to confirm, Esc to cancel
    4. After confirmation, opens external editor to write content; save and exit, and Hunea writes back to the corresponding scope

    It roughly looks like this:

    prompt-create

    What are these sources?

    You can think of the left column as "the system assembly list that the next new session will carry". Common sources include:

    1. Core system prompt
      Hunea's built-in core behavior instructions. Can be overridden by global / project override; when selected, you can press r to restore the default core prompt, which is useful for one-click recovery after messy modifications.

    2. Instructions files

      • Global: AGENTS.md in the data directory (usually ~/.config/hunea/AGENTS.md)
      • Project: AGENTS.md or CLAUDE.md found by searching upward from the workspace/repository (when in the same directory, AGENTS.md is usually prioritized)
    3. Custom prompts
      Your own maintained extra prompt fragments. Project-level usually lives in .hunea/prompts/custom/, and assembly state is recorded by .hunea/prompt-assembly.toml.

    4. Skill discovery / long-lived skills
      Skill discovery fragments, and long-lived injectable skills. This is different from temporary @skill injections that are "only attached for this round" in a session.

    5. Tool guidelines
      Tool usage instructions generated based on currently available tools.

    6. Dynamic environment
      Environment information:

      • Base: more for first-round baseline
      • Change: more for incremental changes in subsequent rounds
        You can check these separately in the right Dynamic tab.

    What does shadowed mean?

    Sources with the same name or conflicting keys may exist at project / global / builtin levels simultaneously. During assembly, only the higher-priority one actually takes effect, and the others are marked shadowed.

    Priority roughly is:

    project > global > builtin

    So you may see an Active item with +1 shadowed, meaning there's another same-key source covered underneath. Expand with Ctrl + E to verify which layer is actually active.

    What happens after closing

    When you press Esc to close /prompt, Hunea tries to commit/persist the working copy from this panel, rather than "just look and discard".

    The success feedback depends on the effective scope:

    CaseCommon feedback
    Mainly affects the next new sessionToast: Prompt updated. Applies to next new session.
    Current session is still empty, can update current session assemblySystem message: Prompt updated for current empty session.

    Note:

    • For the current session that already has conversation content, modifying assembly in /prompt generally won't rewrite the entire historical transcript.
    • To get most changes into the next round completely, the safest approach is: after closing /prompt, use /clear (alias /new) to start a new session.
    • If opening /prompt fails to enter the editing session (e.g., begin fails), the panel may close directly without going through the commit flow.

    Where is configuration persisted

    Common locations for project-level:

    .hunea/prompt-assembly.toml
    .hunea/prompts/
    .hunea/prompts/custom/

    Global side is related to Hunea's data directory (usually ~/.config/hunea/), for example global AGENTS.md, and global prompt assembly state (managed by session store / data directory, not just current workspace files).

    You don't need to manually write all files before using /prompt: many custom items can be created and edited with a / e in the panel, then Hunea writes them back.

    What you want to doMore appropriate command
    Inspect/adjust the system assembly list for the next new session/prompt
    Check how much context is used / remaining/context
    Switch model for current session/models
    Clear current conversation, start a new session/clear (alias /new)
    Copy user/assistant messages from current session/copy

    /context answers "how much budget is left"; /prompt answers "among this budget, what sources make up the system side, and can it be modified".

    Usage tips

    • You just want to confirm "what exactly will the next round carry": open and press p to see the Assembled prompt, which is usually faster than opening one by one.
    • Your project has AGENTS.md / CLAUDE.md but it doesn't seem to take effect: look for instructions in the left column, check if it's missing, disabled, or shadowed by a higher-priority source.
    • You want to add a long-term instruction just for the current repository: switch right to Custom Prompts, a select Project, after writing confirm it appears in the left Active list.
    • After changing order or toggling, preview with p first, then Esc to close and persist; if you need to verify immediately in clean context, then /clear to start a new session.
    • If you can't remember all shortcuts, press ? to see More — easier than guessing footer abbreviations.