Crate caliban_output_styles 

Output styles for the caliban agent harness.

Splices an <output-style name="...">...</output-style> block into the system prompt to nudge the model toward a particular response shape (explanatory commentary, learning-paced prompts with TODO(human) markers, etc.) without touching tools, permissions, or hooks.

See docs/superpowers/specs/2026-05-24-output-styles-design.md and docs/adr/0031-output-styles.md.

Re-exports

pub use learning::IdentityPostProcessor;

pub use learning::LearningPostProcessor;

pub use learning::insert_todo_human_markers;

pub use loader::DiscoveryRoots;

pub use loader::OutputStyleError;

pub use loader::default_roots;

pub use loader::load_one;

pub use loader::load_styles;

pub use loader::select_active;

pub use prefix::OutputStylePrefix;

pub use registry::OutputStylesRegistry;

pub use style::OutputStyle;

pub use style::OutputStyleSource;

Modules

learning

Post-processor for the Learning style.

loader

Filesystem walker + frontmatter parser for output-style .md files.

prefix

OutputStylePrefix — splice the active style’s body into the system prompt.

registry

OutputStylesRegistry — owns the loaded styles and exposes name-based lookup for the binary, the TUI overlay, and the system-prompt splice.

style

OutputStyle struct + frontmatter shape.

Constants

ACTIVE_STYLE_ENV

Environment variable that selects the active output style by name.

Functions

requested_from_env

Read the requested output-style name from the environment.