Module system_prompt 

System prompt construction for the agent loop and channel subsystem. These functions were originally in channels/mod.rs but live here to break a circular dependency between the channels and agent modules.

Structs

BootstrapTruncation

Result of applying the per-file bootstrap cap to one workspace file. Counts at the per-file stage; the whole-prompt budget (finalize_system_prompt) may cut further.

Constants

BOOTSTRAP_FILES

Workspace files injected into every system prompt, unconditionally.

BOOTSTRAP_MAX_CHARS

Maximum characters per injected workspace file (matches OpenClaw default).

COMPACT_BOOTSTRAP_MAX_CHARS

Per-file cap applied to every injected workspace bootstrap file while the agent’s runtime profile has compact_context on (the default). BOOTSTRAP_MAX_CHARS is the cap when it is off. Shared by the turn paths and doctor so they report the same budget.

CONDITIONAL_BOOTSTRAP_FILES

Bootstrap files injected only under conditions: BOOTSTRAP.md only when present, MEMORY.md only when memory is injected. doctor scans both lists.

NATIVE_TOOLS_TASK_FRAMING

NO_TOOLS_TASK_FRAMING

Functions

build_skills_prompt_with_effective_tools

Render only the skills section against an assembled effective tool surface. Context-free callers should continue using crate::skills::skills_to_prompt_with_mode.

build_system_prompt

Build the default system prompt.

build_system_prompt_with_mode

build_system_prompt_with_mode_and_autonomy

build_system_prompt_with_mode_and_effective_tools

Build the system prompt with the effective callable tool names supplied by the turn’s assembled registry. The tool descriptions remain separately filtered for the prompt surface, while skill callable metadata uses this name set as its availability source of truth.

build_system_prompt_with_tool_calls

Like build_system_prompt but accepts show_tool_calls to control whether the system prompt instructs the model to hide tool narration.

finalize_system_prompt

Apply the final model-visible prompt budget in Unicode scalar values.

truncate_bootstrap_content

Trim and cap content exactly as prompt injection does. Returns the text that would be injected and, when the cap applied, the counts.