Module llm 

LLM prompt formatting — template resolution and prompt assembly.

Provides the built-in prompt templates and the helpers needed to build a structured LLM prompt from sanitized content and an optional sanitization report.

Built-in templates

Name	Use case
"troubleshoot"	Incident triage — root cause, event sequence, remediation
"review-config"	Config review — misconfigurations, security concerns, best practices
"review-security"	Security posture — auth, exposure, TLS, CVEs, hardcoded secrets

A filesystem path can be supplied instead of a name; the file’s raw content is used as-is (no substitution is applied to custom templates).

Prompt modes

Inline (format_llm_prompt) — sanitized bytes are embedded directly in <content> blocks. Use when piping output to an LLM without writing files.

Reference (format_llm_prompt_reference) — sanitized files are written to disk and the prompt lists their absolute paths. Use with --output so an agentic LLM can read the files via its own tools.

Example

use sanitize_engine::llm::{format_llm_prompt, LlmEntry};

let entries: Vec<LlmEntry> = vec![
    ("app.log".to_string(), b"INFO start\nERROR disk full\n".to_vec()),
];
let prompt = format_llm_prompt("troubleshoot", &entries, None).unwrap();
assert!(prompt.contains("Root cause"));
assert!(prompt.contains("<content name=\"app.log\">"));

Constants

PROMPT_PREAMBLE

Preamble injected into every built-in template, explaining the sanitization model to the LLM so it does not attempt to recover original values.

TEMPLATE_REVIEW_CONFIG

Built-in template for configuration review.

TEMPLATE_REVIEW_SECURITY

Built-in template for security posture review.

TEMPLATE_TROUBLESHOOT

Built-in template for incident troubleshooting.

Functions

format_llm_prompt

Build a complete LLM prompt from a template, content entries, and an optional sanitization report.

format_llm_prompt_reference

Build a reference-mode LLM prompt: sanitized files are written to disk and the prompt lists their absolute paths for an agentic LLM to read directly.

resolve_llm_template

Resolve a template name or path to its instruction text.

Type Aliases

LlmEntry

A single content entry for the LLM prompt: (label, sanitized_bytes).

LlmPathEntry

A reference entry for the LLM prompt: (input_label, sanitized_output_path).