huskies/server/src/agents/pool/pipeline/completion/server.rs

//! Server-owned completion: runs acceptance gates when an agent process exits normally.
use crate::io::watcher::WatcherEvent;
use crate::slog;
use std::collections::HashMap;
use std::sync::{Arc, Mutex};
use tokio::sync::broadcast;

use super::super::super::super::{AgentEvent, CompletionReport, PipelineStage, pipeline_stage};
use super::super::super::{AgentPool, StoryAgent, composite_key};
use super::super::advance::spawn_pipeline_advance;

/// Server-owned completion: runs acceptance gates when an agent process exits
/// normally, and advances the pipeline based on results.
///
/// This is a **free function** (not a method on `AgentPool`) to break the
/// opaque type cycle that would otherwise arise: `start_agent` → spawned task
/// → server-owned completion → pipeline advance → `start_agent`.
///
/// If the agent already has a completion report (e.g. from a legacy
/// `report_completion` call), this is a no-op to avoid double-running gates.
pub(in crate::agents::pool) async fn run_server_owned_completion(
    agents: &Arc<Mutex<HashMap<String, StoryAgent>>>,
    port: u16,
    story_id: &str,
    agent_name: &str,
    session_id: Option<String>,
    watcher_tx: broadcast::Sender<WatcherEvent>,
) {
    let key = composite_key(story_id, agent_name);

    // Guard: mergemaster agents have their own completion path via
    // start_merge_agent_work / run_merge_pipeline.  Running server-owned gates
    // for a mergemaster would wrongly advance the story to 5_done/ even when
    // no squash merge has occurred (e.g. rate-limited exit before the agent
    // called start_merge_agent_work).  The lifecycle caller is responsible for
    // cleaning up the agent entry and triggering auto-assign.
    if pipeline_stage(agent_name) == PipelineStage::Mergemaster {
        slog!(
            "[agents] run_server_owned_completion skipped for mergemaster \
             '{story_id}:{agent_name}'; mergemaster completion is handled by \
             start_merge_agent_work."
        );
        return;
    }

    // Guard: skip if completion was already recorded (legacy path).
    {
        let lock = match agents.lock() {
            Ok(a) => a,
            Err(_) => return,
        };
        match lock.get(&key) {
            Some(agent) if agent.completion.is_some() => {
                slog!(
                    "[agents] Completion already recorded for '{story_id}:{agent_name}'; \
                     skipping server-owned gates."
                );
                return;
            }
            Some(_) => {}
            None => return,
        }
    }

    // Get worktree path for running gates.
    let worktree_path = {
        let lock = match agents.lock() {
            Ok(a) => a,
            Err(_) => return,
        };
        lock.get(&key)
            .and_then(|a| a.worktree_info.as_ref().map(|wt| wt.path.clone()))
    };

    // Kill any in-flight cargo test processes for this worktree so they don't
    // hold the build lock while gates try to run.
    if let Some(wt_path) = worktree_path.as_ref()
        && let Ok(output) = std::process::Command::new("pgrep")
            .args([
                "-f",
                &format!("--manifest-path {}/Cargo.toml", wt_path.display()),
            ])
            .output()
    {
        let pids = String::from_utf8_lossy(&output.stdout);
        for pid_str in pids.lines() {
            if let Ok(pid) = pid_str.trim().parse::<i32>() {
                crate::slog!(
                    "[agents] Killing stale cargo process (pid {pid}) for '{story_id}' before running gates"
                );
                unsafe {
                    libc::kill(pid, libc::SIGKILL);
                }
            }
        }
    }

    // Run acceptance gates.
    let (gates_passed, gate_output) = if let Some(wt_path) = worktree_path {
        let path = wt_path;
        match tokio::task::spawn_blocking(move || {
            // If the worktree is dirty, check whether committed work survived.
            // An agent crash (e.g. Claude Code CLI's `output.write(&bytes).is_ok()`
            // assertion — bug 645) can leave uncommitted files behind even though
            // the agent already committed valid work.  In that case, stash the
            // dirty files and proceed with gates on the committed code.
            // Uncommitted work is never junk — it may be the next agent session's
            // starting point (bug 651).
            let stashed =
                if let Err(dirty_msg) = crate::agents::gates::check_uncommitted_changes(&path) {
                    if crate::agents::gates::worktree_has_committed_work(&path) {
                        crate::slog!(
                            "[agents] Worktree dirty but committed work exists — \
                         stashing uncommitted changes and proceeding with gates. \
                         Dirty state: {dirty_msg}"
                        );
                        // Stash dirty files so gates run against committed code only.
                        // They will be restored after gates complete.
                        std::process::Command::new("git")
                            .args([
                                "stash",
                                "push",
                                "--include-untracked",
                                "-m",
                                "server-completion-temp",
                            ])
                            .current_dir(&path)
                            .output()
                            .map(|o| {
                                o.status.success()
                                    && !String::from_utf8_lossy(&o.stdout)
                                        .contains("No local changes to save")
                            })
                            .unwrap_or(false)
                    } else {
                        return Ok((false, dirty_msg));
                    }
                } else {
                    false
                };
            // AC5: Fail early if the coder finished with no commits on the feature branch.
            // This prevents empty-diff stories from advancing through QA to merge.
            if !crate::agents::gates::worktree_has_committed_work(&path) {
                if stashed {
                    let _ = std::process::Command::new("git")
                        .args(["stash", "pop"])
                        .current_dir(&path)
                        .output();
                }
                return Ok((
                    false,
                    "Agent exited with no commits on the feature branch. \
                     The agent did not produce any code changes."
                        .to_string(),
                ));
            }
            let result = crate::agents::gates::run_acceptance_gates(&path);
            // Restore stashed uncommitted changes.
            if stashed {
                let _ = std::process::Command::new("git")
                    .args(["stash", "pop"])
                    .current_dir(&path)
                    .output();
            }
            result
        })
        .await
        {
            Ok(Ok(result)) => result,
            Ok(Err(e)) => (false, e),
            Err(e) => (false, format!("Gate check task panicked: {e}")),
        }
    } else {
        (
            false,
            "No worktree path available to run acceptance gates".to_string(),
        )
    };

    slog!(
        "[agents] Server-owned completion for '{story_id}:{agent_name}': gates_passed={gates_passed}"
    );

    // Notify chat transports of the agent completion result.
    let _ = watcher_tx.send(WatcherEvent::AgentCompleted {
        story_id: story_id.to_string(),
        agent_name: agent_name.to_string(),
        success: gates_passed,
    });

    let report = CompletionReport {
        summary: "Agent process exited normally".to_string(),
        gates_passed,
        gate_output,
    };

    // Store completion report, extract data for pipeline advance, then
    // remove the entry so completed agents never appear in list_agents.
    let (tx, project_root_for_advance, wt_path_for_advance, merge_failure_reported_for_advance) = {
        let mut lock = match agents.lock() {
            Ok(a) => a,
            Err(_) => return,
        };
        let agent = match lock.get_mut(&key) {
            Some(a) => a,
            None => return,
        };
        agent.completion = Some(report.clone());
        agent.session_id = session_id.clone();
        let tx = agent.tx.clone();
        let pr = agent.project_root.clone();
        let wt = agent.worktree_info.as_ref().map(|w| w.path.clone());
        let mfr = agent.merge_failure_reported;
        lock.remove(&key);
        (tx, pr, wt, mfr)
    };
    // The completed session's ID is used to resume if gates fail.
    let previous_session_id = session_id.clone();

    // Emit Done so wait_for_agent unblocks.
    let _ = tx.send(AgentEvent::Done {
        story_id: story_id.to_string(),
        agent_name: agent_name.to_string(),
        session_id,
    });

    // Notify WebSocket clients that the agent is gone.
    AgentPool::notify_agent_state_changed(&watcher_tx);

    // Advance the pipeline state machine in a background task.
    spawn_pipeline_advance(
        Arc::clone(agents),
        port,
        story_id,
        agent_name,
        report,
        project_root_for_advance,
        wt_path_for_advance,
        watcher_tx,
        merge_failure_reported_for_advance,
        previous_session_id,
    );
}