// knowledge.rs — agent execution and conversation fragment selection
//
// Agent prompts live in agents/*.agent files, dispatched via defs.rs.
// This module handles:
//   - Agent execution (build prompt → call LLM with tools → log)
//   - Conversation fragment selection (for observation agent)
//
// Agents apply changes via tool calls (poc-memory write/link-add/etc)
// during the LLM call — no action parsing needed.

use super::llm;
use crate::store::{self, Store};

use std::fs;
use std::path::PathBuf;

// ---------------------------------------------------------------------------
// Agent execution
// ---------------------------------------------------------------------------

/// Result of running a single agent.
pub struct AgentResult {
    pub output: String,
    pub node_keys: Vec<String>,
}

/// Run a single agent and return the result (no action application — tools handle that).
pub fn run_and_apply(
    store: &mut Store,
    agent_name: &str,
    batch_size: usize,
    llm_tag: &str,
) -> Result<(), String> {
    run_and_apply_with_log(store, agent_name, batch_size, llm_tag, &|_| {})
}

pub fn run_and_apply_with_log(
    store: &mut Store,
    agent_name: &str,
    batch_size: usize,
    llm_tag: &str,
    log: &dyn Fn(&str),
) -> Result<(), String> {
    run_and_apply_excluded(store, agent_name, batch_size, llm_tag, log, &Default::default())
}

/// Like run_and_apply_with_log but with an in-flight exclusion set.
/// Returns the keys that were processed (for the daemon to track).
pub fn run_and_apply_excluded(
    store: &mut Store,
    agent_name: &str,
    batch_size: usize,
    llm_tag: &str,
    log: &dyn Fn(&str),
    exclude: &std::collections::HashSet<String>,
) -> Result<(), String> {
    let result = run_one_agent_excluded(store, agent_name, batch_size, llm_tag, log, false, exclude)?;

    // Mark conversation segments as mined after successful processing
    if agent_name == "observation" {
        mark_observation_done(&result.node_keys);
    }

    Ok(())
}

/// Run an agent with explicit target keys, bypassing the agent's query.
pub fn run_one_agent_with_keys(
    store: &mut Store,
    agent_name: &str,
    keys: &[String],
    count: usize,
    llm_tag: &str,
    log: &dyn Fn(&str),
    debug: bool,
) -> Result<AgentResult, String> {
    let def = super::defs::get_def(agent_name)
        .ok_or_else(|| format!("no .agent file for {}", agent_name))?;

    log(&format!("targeting: {}", keys.join(", ")));
    let graph = store.build_graph();
    let (prompt, extra_keys) = super::defs::resolve_placeholders(
        &def.prompt, store, &graph, keys, count,
    );
    let mut all_keys: Vec<String> = keys.to_vec();
    all_keys.extend(extra_keys);
    let agent_batch = super::prompts::AgentBatch { prompt, node_keys: all_keys };

    // Record visits eagerly so concurrent agents pick different seeds
    if !agent_batch.node_keys.is_empty() {
        store.record_agent_visits(&agent_batch.node_keys, agent_name).ok();
    }

    run_one_agent_inner(store, agent_name, &def, agent_batch, llm_tag, log, debug)
}

pub fn run_one_agent(
    store: &mut Store,
    agent_name: &str,
    batch_size: usize,
    llm_tag: &str,
    log: &dyn Fn(&str),
    debug: bool,
) -> Result<AgentResult, String> {
    run_one_agent_excluded(store, agent_name, batch_size, llm_tag, log, debug, &Default::default())
}

/// Like run_one_agent but excludes nodes currently being worked on by other agents.
pub fn run_one_agent_excluded(
    store: &mut Store,
    agent_name: &str,
    batch_size: usize,
    llm_tag: &str,
    log: &dyn Fn(&str),
    debug: bool,
    exclude: &std::collections::HashSet<String>,
) -> Result<AgentResult, String> {
    let def = super::defs::get_def(agent_name)
        .ok_or_else(|| format!("no .agent file for {}", agent_name))?;

    log("building prompt");
    let effective_count = def.count.unwrap_or(batch_size);
    let agent_batch = super::defs::run_agent(store, &def, effective_count, exclude)?;

    run_one_agent_inner(store, agent_name, &def, agent_batch, llm_tag, log, debug)
}

fn run_one_agent_inner(
    _store: &mut Store,
    agent_name: &str,
    def: &super::defs::AgentDef,
    agent_batch: super::prompts::AgentBatch,
    _llm_tag: &str,
    log: &dyn Fn(&str),
    debug: bool,
) -> Result<AgentResult, String> {
    let prompt_kb = agent_batch.prompt.len() / 1024;
    let tools_desc = if def.tools.is_empty() { "no tools".into() }
        else { format!("{} tools", def.tools.len()) };
    log(&format!("prompt {}KB, model={}, {}, {} nodes",
        prompt_kb, def.model, tools_desc, agent_batch.node_keys.len()));

    // Guard: reject prompts that would exceed model context.
    // Rough estimate: 1 token ≈ 4 bytes. Reserve 16K tokens for output.
    let max_prompt_bytes = 800_000; // ~200K tokens, leaves room for output
    if agent_batch.prompt.len() > max_prompt_bytes {
        // Log the oversized prompt for debugging
        let oversize_dir = store::memory_dir().join("llm-logs").join("oversized");
        fs::create_dir_all(&oversize_dir).ok();
        let oversize_path = oversize_dir.join(format!("{}-{}.txt",
            agent_name, store::compact_timestamp()));
        let header = format!("=== OVERSIZED PROMPT ===\nagent: {}\nsize: {}KB (max {}KB)\nnodes: {:?}\n\n",
            agent_name, prompt_kb, max_prompt_bytes / 1024, agent_batch.node_keys);
        fs::write(&oversize_path, format!("{}{}", header, agent_batch.prompt)).ok();
        log(&format!("oversized prompt logged to {}", oversize_path.display()));

        return Err(format!(
            "prompt too large: {}KB (max {}KB) — seed nodes may be oversized",
            prompt_kb, max_prompt_bytes / 1024,
        ));
    }
    for key in &agent_batch.node_keys {
        log(&format!("  node: {}", key));
    }

    // Single log file: prompt then response
    let log_dir = store::memory_dir().join("llm-logs").join(agent_name);
    fs::create_dir_all(&log_dir).ok();
    let log_path = log_dir.join(format!("{}.txt", store::compact_timestamp()));
    let prompt_section = format!("=== PROMPT ===\n\n{}\n\n=== CALLING LLM ===\n", agent_batch.prompt);
    fs::write(&log_path, &prompt_section).ok();
    if debug { print!("{}", prompt_section); }
    log(&format!("log: {}", log_path.display()));

    log("calling LLM");
    let output = llm::call_for_def(def, &agent_batch.prompt)?;

    // Append response to same log file
    use std::io::Write;
    let response_section = format!("\n=== RESPONSE ===\n\n{}\n", output);
    if let Ok(mut f) = fs::OpenOptions::new().append(true).open(&log_path) {
        write!(f, "{}", response_section).ok();
    }
    if debug { print!("{}", response_section); }
    log(&format!("response {}KB", output.len() / 1024));

    Ok(AgentResult {
        output,
        node_keys: agent_batch.node_keys,
    })
}

// ---------------------------------------------------------------------------
// Conversation fragment selection
// ---------------------------------------------------------------------------

/// Select conversation fragments (per-segment) for the observation extractor.
/// Uses the transcript-progress.capnp log for dedup — no stub nodes.
/// Does NOT pre-mark segments; caller must call mark_observation_done() after success.
pub fn select_conversation_fragments(n: usize) -> Vec<(String, String)> {
    let projects = crate::config::get().projects_dir.clone();
    if !projects.exists() { return Vec::new(); }

    let store = match crate::store::Store::load() {
        Ok(s) => s,
        Err(_) => return Vec::new(),
    };

    let mut jsonl_files: Vec<PathBuf> = Vec::new();
    if let Ok(dirs) = fs::read_dir(&projects) {
        for dir in dirs.filter_map(|e| e.ok()) {
            if !dir.path().is_dir() { continue; }
            if let Ok(files) = fs::read_dir(dir.path()) {
                for f in files.filter_map(|e| e.ok()) {
                    let p = f.path();
                    if p.extension().map(|x| x == "jsonl").unwrap_or(false) {
                        if let Ok(meta) = p.metadata() {
                            if meta.len() > 50_000 {
                                jsonl_files.push(p);
                            }
                        }
                    }
                }
            }
        }
    }

    // Collect unmined segments across all transcripts
    let mut candidates: Vec<(String, String)> = Vec::new();
    for path in &jsonl_files {
        let path_str = path.to_string_lossy();
        let messages = match super::enrich::extract_conversation(&path_str) {
            Ok(m) => m,
            Err(_) => continue,
        };
        let session_id = path.file_stem()
            .map(|s| s.to_string_lossy().to_string())
            .unwrap_or_else(|| "unknown".into());

        let segments = super::enrich::split_on_compaction(messages);
        for (seg_idx, segment) in segments.into_iter().enumerate() {
            if store.is_segment_mined(&session_id, seg_idx as u32, "observation") {
                continue;
            }
            // Skip segments with too few assistant messages (rate limits, errors)
            let assistant_msgs = segment.iter()
                .filter(|(_, role, _, _)| role == "assistant")
                .count();
            if assistant_msgs < 2 {
                continue;
            }
            // Skip segments that are just rate limit errors
            let has_rate_limit = segment.iter().any(|(_, _, text, _)|
                text.contains("hit your limit") || text.contains("rate limit"));
            if has_rate_limit && assistant_msgs < 3 {
                continue;
            }
            let text = format_segment(&segment);
            if text.len() < 500 {
                continue;
            }
            const CHUNK_SIZE: usize = 50_000;
            const OVERLAP: usize = 10_000;
            if text.len() <= CHUNK_SIZE {
                let id = format!("{}.{}", session_id, seg_idx);
                candidates.push((id, text));
            } else {
                // Split on line boundaries with overlap
                let lines: Vec<&str> = text.lines().collect();
                let mut start_line = 0;
                let mut chunk_idx = 0;
                while start_line < lines.len() {
                    let mut end_line = start_line;
                    let mut size = 0;
                    while end_line < lines.len() && size < CHUNK_SIZE {
                        size += lines[end_line].len() + 1;
                        end_line += 1;
                    }
                    let chunk: String = lines[start_line..end_line].join("\n");
                    let id = format!("{}.{}.{}", session_id, seg_idx, chunk_idx);
                    candidates.push((id, chunk));
                    if end_line >= lines.len() { break; }
                    // Back up by overlap amount for next chunk
                    let mut overlap_size = 0;
                    let mut overlap_start = end_line;
                    while overlap_start > start_line && overlap_size < OVERLAP {
                        overlap_start -= 1;
                        overlap_size += lines[overlap_start].len() + 1;
                    }
                    start_line = overlap_start;
                    chunk_idx += 1;
                }
            }
        }

        if candidates.len() >= n { break; }
    }

    candidates.truncate(n);
    candidates
}

/// Mark observation segments as successfully mined (call AFTER the agent succeeds).
pub fn mark_observation_done(fragment_ids: &[String]) {
    let mut store = match crate::store::Store::load() {
        Ok(s) => s,
        Err(_) => return,
    };
    for id in fragment_ids {
        if let Some((session_id, seg_str)) = id.rsplit_once('.') {
            if let Ok(seg) = seg_str.parse::<u32>() {
                let _ = store.mark_segment_mined(session_id, seg, "observation");
            }
        }
    }
}

/// Format a segment's messages into readable text for the observation agent.
fn format_segment(messages: &[(usize, String, String, String)]) -> String {
    let cfg = crate::config::get();
    let mut fragments = Vec::new();

    for (_, role, text, ts) in messages {
        let min_len = if role == "user" { 5 } else { 10 };
        if text.len() <= min_len { continue; }

        let name = if role == "user" { &cfg.user_name } else { &cfg.assistant_name };
        if ts.is_empty() {
            fragments.push(format!("**{}:** {}", name, text));
        } else {
            fragments.push(format!("**{}** {}: {}", name, &ts[..ts.len().min(19)], text));
        }
    }
    fragments.join("\n\n")
}