Function format_learnings_markdown

atomic-agent/src/learnings.rs:158–183

Format learnings as individual list items grouped by category. Only includes **Repo** and **Workflow** learnings. Code learnings (with file:line references) are stored in the change's unhashed section, not in the context file — they're noisy, they drift with edits, and in a large project they'd grow unbounded. Returns a map of heading → list items so callers can merge into existing headings inst

(learnings: &Learnings)

/// - cargo test --lib is faster for auth changes
/// ```
pub fn format_learnings_markdown(learnings: &Learnings) -> String {
    let mut out = String::new();

    if !learnings.repo.is_empty() {
        out.push_str("### Repo\n");
        for item in &learnings.repo {
            out.push_str(&format!("- {}\n", item));
        }
        out.push('\n');
    }

    // Code learnings are intentionally NOT written to the context file.
    // They live in the change's unhashed section where the server can
    // render them with CRDT graph anchoring. The context file only gets
    // the durable, location-independent knowledge.

    if !learnings.workflow.is_empty() {
        out.push_str("### Workflow\n");
        for item in &learnings.workflow {
            out.push_str(&format!("- {}\n", item));
        }
        out.push('\n');
    }

    out
}

/// Count how many learnings would be written to the context file.
///