diff --git a/.beans/claudbg-76fy--implement-jsonl-session-file-reader.md b/.beans/claudbg-76fy--implement-jsonl-session-file-reader.md
index 00ad2f4..6fb0c2f 100644
--- a/.beans/claudbg-76fy--implement-jsonl-session-file-reader.md
+++ b/.beans/claudbg-76fy--implement-jsonl-session-file-reader.md
@@ -1,10 +1,11 @@
 ---
 # claudbg-76fy
 title: Implement JSONL session file reader
-status: todo
+status: in-progress
 type: task
+priority: normal
 created_at: 2026-03-27T19:39:15Z
-updated_at: 2026-03-27T19:39:15Z
+updated_at: 2026-03-28T05:45:28Z
 parent: claudbg-mztt
 ---
 
diff --git a/.beans/claudbg-g5uv--session-file-discovery.md b/.beans/claudbg-g5uv--session-file-discovery.md
index 6107d59..9dbbe03 100644
--- a/.beans/claudbg-g5uv--session-file-discovery.md
+++ b/.beans/claudbg-g5uv--session-file-discovery.md
@@ -1,10 +1,11 @@
 ---
 # claudbg-g5uv
 title: Session file discovery
-status: todo
+status: in-progress
 type: task
+priority: normal
 created_at: 2026-03-27T19:39:15Z
-updated_at: 2026-03-27T19:39:15Z
+updated_at: 2026-03-28T05:43:18Z
 parent: claudbg-mztt
 ---
 
diff --git a/.beans/claudbg-jupi--sub-agent-file-discovery.md b/.beans/claudbg-jupi--sub-agent-file-discovery.md
index 30e9158..91e7796 100644
--- a/.beans/claudbg-jupi--sub-agent-file-discovery.md
+++ b/.beans/claudbg-jupi--sub-agent-file-discovery.md
@@ -1,10 +1,11 @@
 ---
 # claudbg-jupi
 title: Sub-agent file discovery
-status: todo
+status: in-progress
 type: task
+priority: normal
 created_at: 2026-03-27T19:39:15Z
-updated_at: 2026-03-27T19:39:15Z
+updated_at: 2026-03-28T05:45:22Z
 parent: claudbg-mztt
 ---
 
diff --git a/src/lib.rs b/src/lib.rs
index 9ca3cd8..78e6d57 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -6,6 +6,7 @@ pub mod cli;
 pub mod error;
 pub mod models;
 // pub mod output;
+pub mod parser;
 pub mod util;
 
 #[cfg(test)]
diff --git a/src/parser/discovery.rs b/src/parser/discovery.rs
new file mode 100644
index 0000000..ada1d34
--- /dev/null
+++ b/src/parser/discovery.rs
@@ -0,0 +1,451 @@
+//! Discovery of Claude Code session and sub-agent JSONL files on disk.
+
+use std::io::{BufRead, BufReader};
+use std::path::{Path, PathBuf};
+
+use chrono::{DateTime, Utc};
+
+use crate::models::session::RawEntry;
+
+/// Reference to a discovered session file on disk.
+#[derive(Debug, Clone)]
+pub struct SessionRef {
+    /// The session UUID (filename stem, without `.jsonl`).
+    pub session_id: String,
+    /// The project path recovered from the JSONL `cwd` field.
+    /// `None` if the file could not be parsed to find a `cwd`.
+    pub project_path: Option<String>,
+    /// Absolute path to the `.jsonl` file.
+    pub file_path: PathBuf,
+    /// Last-modified time of the file.
+    pub modified_at: DateTime<Utc>,
+}
+
+/// Reference to a discovered sub-agent JSONL file.
+#[derive(Debug, Clone)]
+pub struct AgentRef {
+    /// The agent's UUID (from filename: `agent-{uuid}.jsonl`).
+    pub agent_id: String,
+    /// The parent session UUID (read from first line of the agent JSONL).
+    pub session_id: String,
+    /// Agent type, from `agent-{id}.meta.json` if present.
+    pub agent_type: Option<String>,
+    /// Absolute path to the agent's `.jsonl` file.
+    pub file_path: PathBuf,
+    /// Last-modified time of the file.
+    pub modified_at: DateTime<Utc>,
+}
+
+/// Resolve the `~/.claude/projects/` directory path.
+///
+/// Uses `HOME` environment variable to expand `~`.
+fn claude_projects_dir() -> Option<PathBuf> {
+    let home = std::env::var("HOME").ok()?;
+    Some(PathBuf::from(home).join(".claude").join("projects"))
+}
+
+/// Read the first non-empty line of a file and attempt to extract `cwd` from it.
+///
+/// Returns `None` if the file cannot be read or the first line cannot be parsed.
+fn read_cwd_from_first_line(path: &Path) -> Option<String> {
+    let file = std::fs::File::open(path).ok()?;
+    let reader = BufReader::new(file);
+    for line in reader.lines() {
+        let line = line.ok()?;
+        let trimmed = line.trim();
+        if trimmed.is_empty() {
+            continue;
+        }
+        if let Ok(entry) = serde_json::from_str::<RawEntry>(trimmed)
+            && entry.cwd.is_some()
+        {
+            return entry.cwd;
+        }
+        // Return after first non-empty line regardless.
+        break;
+    }
+    None
+}
+
+/// Discover all session JSONL files under `~/.claude/projects/`.
+///
+/// Walks one level deep: for each project subdirectory, collects `*.jsonl` files
+/// while skipping `subagents/` subdirectories.  For each file the session ID is
+/// extracted from the filename stem and the project path from the first parseable
+/// JSONL line's `cwd` field.
+///
+/// Files that cannot be read or stat'd are skipped with a warning to `stderr`.
+pub fn discover_sessions() -> crate::error::Result<Vec<SessionRef>> {
+    let projects_dir = match claude_projects_dir() {
+        Some(d) => d,
+        None => {
+            eprintln!("claudbg: could not determine HOME directory");
+            return Ok(vec![]);
+        }
+    };
+
+    if !projects_dir.exists() {
+        return Ok(vec![]);
+    }
+
+    let mut sessions = Vec::new();
+
+    let project_entries = match std::fs::read_dir(&projects_dir) {
+        Ok(e) => e,
+        Err(err) => {
+            eprintln!(
+                "claudbg: could not read {}: {}",
+                projects_dir.display(),
+                err
+            );
+            return Ok(vec![]);
+        }
+    };
+
+    for proj_entry in project_entries {
+        let proj_entry = match proj_entry {
+            Ok(e) => e,
+            Err(err) => {
+                eprintln!("claudbg: error reading project dir entry: {err}");
+                continue;
+            }
+        };
+
+        let proj_path = proj_entry.path();
+        if !proj_path.is_dir() {
+            continue;
+        }
+
+        let file_entries = match std::fs::read_dir(&proj_path) {
+            Ok(e) => e,
+            Err(err) => {
+                eprintln!("claudbg: could not read {}: {}", proj_path.display(), err);
+                continue;
+            }
+        };
+
+        for file_entry in file_entries {
+            let file_entry = match file_entry {
+                Ok(e) => e,
+                Err(err) => {
+                    eprintln!("claudbg: error reading file entry: {err}");
+                    continue;
+                }
+            };
+
+            let file_path = file_entry.path();
+
+            // Skip subdirectories (e.g. subagents/).
+            if file_path.is_dir() {
+                continue;
+            }
+
+            // Only process .jsonl files.
+            if file_path.extension().and_then(|e| e.to_str()) != Some("jsonl") {
+                continue;
+            }
+
+            // Extract session_id from filename stem.
+            let session_id = match file_path.file_stem().and_then(|s| s.to_str()) {
+                Some(s) => s.to_string(),
+                None => {
+                    eprintln!(
+                        "claudbg: could not extract session ID from {}",
+                        file_path.display()
+                    );
+                    continue;
+                }
+            };
+
+            // Get last-modified time.
+            let modified_at = match file_entry.metadata() {
+                Ok(meta) => match meta.modified() {
+                    Ok(t) => DateTime::<Utc>::from(t),
+                    Err(err) => {
+                        eprintln!(
+                            "claudbg: could not read mtime for {}: {}",
+                            file_path.display(),
+                            err
+                        );
+                        continue;
+                    }
+                },
+                Err(err) => {
+                    eprintln!("claudbg: could not stat {}: {}", file_path.display(), err);
+                    continue;
+                }
+            };
+
+            // Try to extract cwd from first JSONL line.
+            let project_path = read_cwd_from_first_line(&file_path);
+
+            sessions.push(SessionRef {
+                session_id,
+                project_path,
+                file_path,
+                modified_at,
+            });
+        }
+    }
+
+    Ok(sessions)
+}
+
+/// Discover all sub-agent runs for a session given the session's JSONL file path.
+///
+/// Looks for `subagents/agent-*.jsonl` files in the same directory as
+/// `session_file`.
+pub fn discover_agents_for_session(session_file: &Path) -> crate::error::Result<Vec<AgentRef>> {
+    let parent = match session_file.parent() {
+        Some(p) => p,
+        None => return Ok(vec![]),
+    };
+    let subagents_dir = parent.join("subagents");
+    collect_agents_in_dir(&subagents_dir, None)
+}
+
+/// Discover all sub-agent runs across all sessions.
+///
+/// Walks `~/.claude/projects/` and collects `agent-*.jsonl` files from every
+/// `subagents/` subdirectory found.
+pub fn discover_all_agents() -> crate::error::Result<Vec<AgentRef>> {
+    let projects_dir = match claude_projects_dir() {
+        Some(d) => d,
+        None => {
+            eprintln!("claudbg: could not determine HOME directory");
+            return Ok(vec![]);
+        }
+    };
+
+    if !projects_dir.exists() {
+        return Ok(vec![]);
+    }
+
+    let mut agents = Vec::new();
+
+    let project_entries = match std::fs::read_dir(&projects_dir) {
+        Ok(e) => e,
+        Err(err) => {
+            eprintln!(
+                "claudbg: could not read {}: {}",
+                projects_dir.display(),
+                err
+            );
+            return Ok(vec![]);
+        }
+    };
+
+    for proj_entry in project_entries {
+        let proj_entry = match proj_entry {
+            Ok(e) => e,
+            Err(err) => {
+                eprintln!("claudbg: error reading project dir entry: {err}");
+                continue;
+            }
+        };
+
+        let proj_path = proj_entry.path();
+        if !proj_path.is_dir() {
+            continue;
+        }
+
+        let subagents_dir = proj_path.join("subagents");
+        match collect_agents_in_dir(&subagents_dir, None) {
+            Ok(mut found) => agents.append(&mut found),
+            Err(err) => {
+                eprintln!(
+                    "claudbg: error collecting agents in {}: {}",
+                    subagents_dir.display(),
+                    err
+                );
+            }
+        }
+    }
+
+    Ok(agents)
+}
+
+/// Collect all `agent-*.jsonl` files from a single `subagents/` directory.
+///
+/// `filter_session_id` optionally restricts results to agents whose first JSONL
+/// line's `session_id` field matches the given value.
+fn collect_agents_in_dir(
+    subagents_dir: &Path,
+    filter_session_id: Option<&str>,
+) -> crate::error::Result<Vec<AgentRef>> {
+    if !subagents_dir.exists() {
+        return Ok(vec![]);
+    }
+
+    let mut agents = Vec::new();
+
+    let entries = match std::fs::read_dir(subagents_dir) {
+        Ok(e) => e,
+        Err(err) => {
+            eprintln!(
+                "claudbg: could not read {}: {}",
+                subagents_dir.display(),
+                err
+            );
+            return Ok(vec![]);
+        }
+    };
+
+    for entry in entries {
+        let entry = match entry {
+            Ok(e) => e,
+            Err(err) => {
+                eprintln!("claudbg: error reading subagent entry: {err}");
+                continue;
+            }
+        };
+
+        let file_path = entry.path();
+        if file_path.is_dir() {
+            continue;
+        }
+
+        // Only process agent-*.jsonl files.
+        let file_name = match file_path.file_name().and_then(|n| n.to_str()) {
+            Some(n) => n.to_string(),
+            None => continue,
+        };
+
+        if !file_name.starts_with("agent-") || !file_name.ends_with(".jsonl") {
+            continue;
+        }
+
+        // Extract agent_id: strip "agent-" prefix and ".jsonl" suffix.
+        let agent_id = file_name
+            .strip_prefix("agent-")
+            .and_then(|s| s.strip_suffix(".jsonl"))
+            .unwrap_or(&file_name)
+            .to_string();
+
+        // Get last-modified time.
+        let modified_at = match entry.metadata() {
+            Ok(meta) => match meta.modified() {
+                Ok(t) => DateTime::<Utc>::from(t),
+                Err(err) => {
+                    eprintln!(
+                        "claudbg: could not read mtime for {}: {}",
+                        file_path.display(),
+                        err
+                    );
+                    continue;
+                }
+            },
+            Err(err) => {
+                eprintln!("claudbg: could not stat {}: {}", file_path.display(), err);
+                continue;
+            }
+        };
+
+        // Read session_id from first line of the agent file.
+        let session_id = read_session_id_from_first_line(&file_path).unwrap_or_default();
+
+        // Filter by session_id if requested.
+        if let Some(filter) = filter_session_id
+            && session_id != filter
+        {
+            continue;
+        }
+
+        // Try to read agent_type from meta.json.
+        let meta_path = subagents_dir.join(format!("agent-{agent_id}.meta.json"));
+        let agent_type = read_agent_type_from_meta(&meta_path);
+
+        agents.push(AgentRef {
+            agent_id,
+            session_id,
+            agent_type,
+            file_path,
+            modified_at,
+        });
+    }
+
+    Ok(agents)
+}
+
+/// Read the `session_id` from the first non-empty line of an agent JSONL file.
+fn read_session_id_from_first_line(path: &Path) -> Option<String> {
+    let file = std::fs::File::open(path).ok()?;
+    let reader = BufReader::new(file);
+    for line in reader.lines() {
+        let line = line.ok()?;
+        let trimmed = line.trim();
+        if trimmed.is_empty() {
+            continue;
+        }
+        if let Ok(entry) = serde_json::from_str::<RawEntry>(trimmed)
+            && entry.session_id.is_some()
+        {
+            return entry.session_id;
+        }
+        break;
+    }
+    None
+}
+
+/// Try to read `agent_type` from an `agent-{id}.meta.json` file.
+///
+/// Expects a JSON object with an `"agent_type"` string field.
+fn read_agent_type_from_meta(meta_path: &Path) -> Option<String> {
+    let content = std::fs::read_to_string(meta_path).ok()?;
+    let value: serde_json::Value = serde_json::from_str(&content).ok()?;
+    value
+        .get("agent_type")
+        .and_then(|v| v.as_str())
+        .map(|s| s.to_string())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// `SessionRef` implements `Debug` and `Clone`.
+    #[test]
+    fn session_ref_debug_clone() {
+        let sr = SessionRef {
+            session_id: "abc123".to_string(),
+            project_path: Some("/home/user/project".to_string()),
+            file_path: PathBuf::from("/tmp/abc123.jsonl"),
+            modified_at: DateTime::<Utc>::from(std::time::SystemTime::UNIX_EPOCH),
+        };
+        let cloned = sr.clone();
+        assert_eq!(cloned.session_id, sr.session_id);
+        // Ensure Debug works without panic.
+        let _ = format!("{sr:?}");
+    }
+
+    /// `AgentRef` implements `Debug` and `Clone`.
+    #[test]
+    fn agent_ref_debug_clone() {
+        let ar = AgentRef {
+            agent_id: "def456".to_string(),
+            session_id: "abc123".to_string(),
+            agent_type: Some("TaskAgent".to_string()),
+            file_path: PathBuf::from("/tmp/agent-def456.jsonl"),
+            modified_at: DateTime::<Utc>::from(std::time::SystemTime::UNIX_EPOCH),
+        };
+        let cloned = ar.clone();
+        assert_eq!(cloned.agent_id, ar.agent_id);
+        let _ = format!("{ar:?}");
+    }
+
+    /// `discover_sessions()` does not panic and returns `Ok` even if
+    /// `~/.claude/projects/` does not exist.
+    #[test]
+    fn discover_sessions_does_not_panic() {
+        let result = discover_sessions();
+        // Should always succeed (empty or non-empty).
+        assert!(result.is_ok());
+    }
+
+    /// `discover_all_agents()` does not panic and returns `Ok`.
+    #[test]
+    fn discover_all_agents_does_not_panic() {
+        let result = discover_all_agents();
+        assert!(result.is_ok());
+    }
+}
diff --git a/src/parser/mod.rs b/src/parser/mod.rs
new file mode 100644
index 0000000..e262ba8
--- /dev/null
+++ b/src/parser/mod.rs
@@ -0,0 +1,3 @@
+//! JSONL session file discovery and reading.
+pub mod discovery;
+pub mod reader;
diff --git a/src/parser/reader.rs b/src/parser/reader.rs
new file mode 100644
index 0000000..7af387d
--- /dev/null
+++ b/src/parser/reader.rs
@@ -0,0 +1,115 @@
+//! Async JSONL session file reader.
+
+use std::path::Path;
+
+use crate::error::Result;
+use crate::models::session::RawEntry;
+
+/// Reads a JSONL session file and returns all successfully-parsed entries.
+///
+/// - Skips empty lines.
+/// - Skips lines that fail to parse (logs a warning with `eprintln!`).
+/// - Uses tokio async I/O with `BufReader` for efficiency.
+/// - If the file path does not exist, returns `Err(AppError::Io(...))`.
+pub async fn read_session_file(path: &Path) -> Result<Vec<RawEntry>> {
+    use tokio::io::AsyncBufReadExt;
+
+    let file = tokio::fs::File::open(path)
+        .await
+        .map_err(crate::error::AppError::Io)?;
+    let reader = tokio::io::BufReader::new(file);
+    let mut lines = reader.lines();
+
+    let mut entries = Vec::new();
+
+    loop {
+        let line = lines
+            .next_line()
+            .await
+            .map_err(crate::error::AppError::Io)?;
+        match line {
+            None => break,
+            Some(line) => {
+                let trimmed = line.trim();
+                if trimmed.is_empty() {
+                    continue;
+                }
+                match serde_json::from_str::<RawEntry>(trimmed) {
+                    Ok(entry) => entries.push(entry),
+                    Err(err) => {
+                        eprintln!("claudbg: skipping unparseable JSONL line: {err}");
+                    }
+                }
+            }
+        }
+    }
+
+    Ok(entries)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use std::io::Write;
+
+    /// Helper: create a temp file with the given content, run the async reader.
+    async fn read_temp(content: &str) -> Result<Vec<RawEntry>> {
+        let dir = std::env::temp_dir();
+        let path = dir.join(format!(
+            "claudbg_test_{}.jsonl",
+            std::time::SystemTime::now()
+                .duration_since(std::time::UNIX_EPOCH)
+                .unwrap_or_default()
+                .subsec_nanos()
+        ));
+        {
+            let mut f = std::fs::File::create(&path).expect("create temp file");
+            f.write_all(content.as_bytes()).expect("write temp file");
+        }
+        let result = read_session_file(&path).await;
+        let _ = std::fs::remove_file(&path);
+        result
+    }
+
+    /// Two valid JSON lines → two entries returned.
+    #[tokio::test]
+    async fn two_valid_lines() {
+        let content = concat!(
+            r#"{"type":"user","session_id":"abc"}"#,
+            "\n",
+            r#"{"type":"assistant","session_id":"abc"}"#,
+            "\n"
+        );
+        let entries = read_temp(content).await.expect("should succeed");
+        assert_eq!(entries.len(), 2);
+    }
+
+    /// One valid + one invalid JSON line → one entry returned (invalid skipped).
+    #[tokio::test]
+    async fn one_valid_one_invalid() {
+        let content = concat!(
+            r#"{"type":"user","session_id":"abc"}"#,
+            "\n",
+            "THIS IS NOT JSON\n"
+        );
+        let entries = read_temp(content).await.expect("should succeed");
+        assert_eq!(entries.len(), 1);
+        assert_eq!(entries[0].entry_type.as_deref(), Some("user"));
+    }
+
+    /// Empty file → empty vec returned.
+    #[tokio::test]
+    async fn empty_file_returns_empty_vec() {
+        let entries = read_temp("").await.expect("should succeed");
+        assert!(entries.is_empty());
+    }
+
+    /// Nonexistent path → returns `Err`.
+    #[tokio::test]
+    async fn nonexistent_path_returns_err() {
+        let path = std::path::PathBuf::from("/tmp/claudbg_nonexistent_8675309.jsonl");
+        let result = read_session_file(&path).await;
+        assert!(result.is_err());
+        assert!(matches!(result, Err(crate::error::AppError::Io(_))));
+    }
+}