feat: CSA-lite Phase 1 — caller session auto-detection + fork-from-caller (#1432)

Cargo.toml

@@ -20,7 +20,7 @@ default-members = [
 resolver = "2"
 
 [workspace.package]
-version = "0.1.732"
+version = "0.1.733"
 edition = "2024"
 rust-version = "1.88"
 license = "Apache-2.0"

crates/cli-sub-agent/src/cli.rs

@@ -96,17 +96,22 @@ pub enum Commands {
         #[arg(long, value_name = "SESSION_ID", value_parser = validate_ulid)]
         inline_context_from_review_session: Option<String>,
         /// Resume existing session (ULID or prefix match) [DEPRECATED: use --fork-from]
-        #[arg(short, long, conflicts_with_all = ["last", "fork_from", "fork_last"])]
+        #[arg(short, long, conflicts_with_all = ["last", "fork_from", "fork_last", "fork_from_caller"])]
         session: Option<String>,
         /// Resume the most recent session for this project [DEPRECATED: use --fork-last]
-        #[arg(long, conflicts_with_all = ["session", "ephemeral", "fork_from", "fork_last"])]
+        #[arg(long, conflicts_with_all = ["session", "ephemeral", "fork_from", "fork_last", "fork_from_caller"])]
         last: bool,
         /// Fork from a specific session (ULID or prefix match)
-        #[arg(long, conflicts_with_all = ["session", "last", "fork_last", "ephemeral"])]
+        #[arg(long, conflicts_with_all = ["session", "last", "fork_last", "fork_from_caller", "ephemeral"])]
         fork_from: Option<String>,
         /// Fork the most recent session for this project
-        #[arg(long, conflicts_with_all = ["session", "last", "fork_from", "ephemeral"])]
+        #[arg(long, conflicts_with_all = ["session", "last", "fork_from", "fork_from_caller", "ephemeral"])]
         fork_last: bool,
+        /// Fork from the auto-detected caller's Claude conversation (CSA-lite, #1432).
+        /// Reads `CLAUDE_SESSION_ID` or the most recent Claude thread on disk,
+        /// extracts a budgeted conversation prefix, and seeds the new session with it.
+        #[arg(long, conflicts_with_all = ["session", "last", "fork_from", "fork_last", "ephemeral"])]
+        fork_from_caller: bool,
         /// Human-readable description for a new session
         #[arg(short, long)]
         description: Option<String>,

crates/cli-sub-agent/src/goal_loop.rs

@@ -75,6 +75,7 @@ pub(crate) struct GoalRunRequest {
     pub(crate) last: bool,
     pub(crate) fork_from: Option<String>,
     pub(crate) fork_last: bool,
+    pub(crate) fork_from_caller: bool,
     pub(crate) description: Option<String>,
     pub(crate) fork_call: bool,
     pub(crate) return_to: Option<String>,
@@ -138,6 +139,7 @@ pub(crate) async fn handle_run_or_goal(request: GoalRunRequest) -> Result<i32> {
         request.last,
         request.fork_from,
         request.fork_last,
+        request.fork_from_caller,
         request.description,
         request.fork_call,
         request.return_to,
@@ -293,6 +295,7 @@ async fn run_goal_iteration(
         first_iteration && request.last,
         first_iteration.then(|| request.fork_from.clone()).flatten(),
         first_iteration && request.fork_last,
+        first_iteration && request.fork_from_caller,
         request.description.clone(),
         request.fork_call,
         request.return_to.clone(),

crates/cli-sub-agent/src/main.rs

@@ -68,6 +68,7 @@ mod review_prior_rounds;
 mod review_routing;
 mod review_session_findings;
 mod run_cmd;
+mod run_cmd_caller_fork;
 mod run_cmd_daemon;
 mod run_cmd_fork;
 mod run_cmd_post;
@@ -80,6 +81,7 @@ mod self_update;
 mod session_cmds;
 mod session_cmds_daemon;
 mod session_cmds_result;
+mod session_cmds_result_measure;
 mod session_dispatch;
 mod session_guard;
 mod session_observability;
@@ -243,10 +245,8 @@ async fn run() -> Result<()> {
                 });
 
         if auto_upgrade {
-            use std::time::Duration;
-
             let mut success = false;
-            let mut delay = Duration::from_secs(1);
+            let mut delay = std::time::Duration::from_secs(1);
 
             for attempt in 0..3 {
                 let result = tokio::process::Command::new("weave")
@@ -281,14 +281,6 @@ async fn run() -> Result<()> {
 
     let legacy_xdg_paths = csa_config::paths::legacy_paths_requiring_migration();
     if !legacy_xdg_paths.is_empty() {
-        for path in &legacy_xdg_paths {
-            tracing::debug!(
-                label = path.label,
-                legacy = %path.legacy_path.display(),
-                new = %path.new_path.display(),
-                "legacy XDG path detected, auto-migrating"
-            );
-        }
         match csa_config::migrate::run_xdg_migration() {
             Ok(()) => {
                 tracing::debug!(
@@ -320,6 +312,7 @@ async fn run() -> Result<()> {
             last,
             fork_from,
             fork_last,
+            fork_from_caller,
             description,
             fork_call,
             return_to,
@@ -404,6 +397,7 @@ async fn run() -> Result<()> {
                 last,
                 fork_from,
                 fork_last,
+                fork_from_caller,
                 description,
                 fork_call,
                 return_to,

crates/cli-sub-agent/src/pipeline_post_exec.rs

@@ -691,6 +691,12 @@ fn update_cumulative_tokens(session: &mut MetaSessionState, token_usage: Option<
     cumulative.estimated_cost_usd = Some(
         cumulative.estimated_cost_usd.unwrap_or(0.0) + new_usage.estimated_cost_usd.unwrap_or(0.0),
     );
+    // Accumulate cache-read tokens only when the new usage reports a value;
+    // missing fields must not zero out a previously recorded total.
+    if let Some(new_cache_read) = new_usage.cache_read_input_tokens {
+        cumulative.cache_read_input_tokens =
+            Some(cumulative.cache_read_input_tokens.unwrap_or(0) + new_cache_read);
+    }
 
     // Update token budget tracking
     if let Some(ref mut budget) = session.token_budget {

crates/cli-sub-agent/src/run_cmd.rs

@@ -114,6 +114,7 @@ impl SubagentRunConfig {
             false,
             None,
             false,
+            false,
             None,
             false,
             None,

crates/cli-sub-agent/src/run_cmd_attempt.rs

@@ -81,7 +81,7 @@ pub(crate) async fn execute_run_loop(request: RunLoopRequest<'_>) -> Result<RunL
     );
     let mut executed_session_id: Option<String> = None;
     let mut pre_created_fork_session_id: Option<String> = None;
-    let mut fork_resolution: Option<ForkResolution> = None;
+    let mut fork_resolution: Option<ForkResolution> = request.caller_fork_resolution;
     let mut is_fork = request.is_fork;
     let mut failover_context_addendum: Option<String> = None;
     let mut is_auto_seed_fork = request.is_auto_seed_fork;

crates/cli-sub-agent/src/run_cmd_attempt_types.rs

@@ -44,6 +44,11 @@ pub(crate) struct RunLoopRequest<'a> {
     pub(crate) run_started_at: Instant,
     pub(crate) is_fork: bool,
     pub(crate) is_auto_seed_fork: bool,
+    /// Pre-resolved fork from `--fork-from-caller` (CSA-lite, #1432).
+    /// When present, supplies the initial fork_resolution before the loop
+    /// runs `resolve_fork()`; downstream prepend-to-prompt path picks up
+    /// the extracted caller conversation prefix.
+    pub(crate) caller_fork_resolution: Option<ForkResolution>,
     pub(crate) ephemeral: bool,
     pub(crate) fork_call: bool,
     pub(crate) session_arg: Option<String>,