fix: address final review (path traversal, sync races, MCP limits)

Findings from a Codex (gpt-6.1-sol, xhigh) review of v0.2.4..HEAD, each
verified in the code before fixing:

- high: ticker symbols reached cache file paths unchecked, so a symbol
  like `../../REPORT` read (and deleted, if unparseable) .json files
  outside the cache, reachable through MCP get_quotes. resolve_symbol now
  accepts only ticker characters, and cache keys are encoded as single
  safe path components.
- high: sync decided before downloading and replaced the database
  without rechecking; an import that landed meanwhile was erased. It now
  rechecks the local state before installing, and idx-mcp runs
  ownership_sync exclusively with ownership reads.
- replaces_sqlite_sha256 never downgrades history: the client requires
  the rebuild to keep at least the local release count, and the
  publisher lists a replaced build only then.
- cache writes are atomic (temp file + rename), so concurrent readers
  never see, and delete, a partial entry.
- idx-mcp caps symbols (20, deduplicated) and limits, and runs at most 4
  tool calls at once.
- each Yahoo client gets its own cookie jar, paired with its crumb.
- ZIP entries (XLSX, KSEI archives) are read with a 256 MiB bound;
  non-ASCII archive dates and huge percentages are parse errors instead
  of panics or overflow.
- agent-smoke --keep deletes every copied agent config, not only auth
  files.

CLI goldens unchanged; the MCP tool golden changes only to state the
new caps.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

crates/idx-core/src/api/mod.rs

@@ -163,6 +163,19 @@ pub fn resolve_symbol(symbol: &str, exchange: &str) -> Result<String, IdxError>
             "ticker symbol cannot be empty".into(),
         ));
     }
+    // Tickers name cache files and provider URLs: letters, digits, and the
+    // separators real symbols use (BBCA.JK, BRK-B, ^JKSE), nothing else.
+    if trimmed.len() > 32
+        || trimmed.contains("..")
+        || !trimmed
+            .chars()
+            .all(|ch| ch.is_ascii_alphanumeric() || matches!(ch, '.' | '-' | '^' | '='))
+    {
+        return Err(IdxError::InvalidInput(format!(
+            "invalid ticker symbol '{}'",
+            symbol.trim()
+        )));
+    }
     if let Some((base, suffix)) = trimmed.rsplit_once('.')
         && !base.is_empty()
         && !suffix.is_empty()
@@ -301,6 +314,26 @@ mod tests {
         assert_eq!(resolve_symbol("BBCA", "JK").unwrap(), "BBCA.JK");
     }
 
+    #[test]
+    fn rejects_symbols_that_are_not_tickers() {
+        for bad in [
+            "../../REPORT",
+            "a/b",
+            "a\\b",
+            "BB CA",
+            "BBCA..JK",
+            "BBCA\u{0}",
+            &"A".repeat(40),
+        ] {
+            assert!(
+                matches!(resolve_symbol(bad, "JK"), Err(IdxError::InvalidInput(_))),
+                "{bad:?} accepted"
+            );
+        }
+        assert_eq!(resolve_symbol("^jkse", "JK").unwrap(), "^JKSE.JK");
+        assert_eq!(resolve_symbol("BRK-B.US", "JK").unwrap(), "BRK-B.US");
+    }
+
     #[test]
     fn selected_provider_exposes_expected_capabilities_for_msn() {
         let provider = build_selected_provider(ProviderKind::Msn, false, false);

crates/idx-core/src/api/yahoo/client.rs

@@ -2,6 +2,7 @@ use std::path::{Path, PathBuf};
 use std::process::Command;
 use std::process::Stdio;
 use std::sync::Mutex;
+use std::sync::atomic::{AtomicU64, Ordering};
 use std::time::Duration;
 
 use crate::api::types::{Interval, Period};
@@ -19,6 +20,9 @@ const CRUMB_FETCH_URL: &str = "https://query1.finance.yahoo.com/v1/test/getcrumb
 pub(super) struct YahooClient {
     agent: ureq::Agent,
     crumb: Mutex<Option<String>>,
+    /// One jar per client, paired with its crumb: clients in the same process
+    /// (concurrent MCP calls) must not overwrite each other's cookies.
+    cookie_jar: PathBuf,
 }
 
 impl YahooClient {
@@ -29,9 +33,16 @@ impl YahooClient {
             .build()
             .into();
 
+        static NEXT_JAR: AtomicU64 = AtomicU64::new(0);
+        let cookie_jar = std::env::temp_dir().join(format!(
+            "idx_yf_{}_{}.txt",
+            std::process::id(),
+            NEXT_JAR.fetch_add(1, Ordering::Relaxed)
+        ));
         Self {
             agent,
             crumb: Mutex::new(None),
+            cookie_jar,
         }
     }
 
@@ -49,10 +60,6 @@ impl YahooClient {
         )
     }
 
-    fn cookie_jar_path() -> PathBuf {
-        std::env::temp_dir().join(format!("idx_yf_{}.txt", std::process::id()))
-    }
-
     pub(super) fn parse_crumb_body(raw: &str) -> Result<String, IdxError> {
         let crumb = raw.trim();
         if crumb.is_empty() {
@@ -123,7 +130,7 @@ impl YahooClient {
     fn fetch_crumb_via_curl(&self) -> Result<String, IdxError> {
         let binary = curl_impersonate::chrome_curl_binary()?;
 
-        let cookie_jar = Self::cookie_jar_path();
+        let cookie_jar = &self.cookie_jar;
         let cookie_jar_str = cookie_jar.to_str().ok_or_else(|| {
             IdxError::Io(format!(
                 "failed to encode Yahoo cookie jar path {}",
@@ -235,7 +242,7 @@ impl YahooClient {
     ) -> Result<QuoteSummaryResponse, IdxError> {
         for auth_attempt in 0..2 {
             let crumb = self.get_or_init_crumb()?;
-            let cookie_header = match Self::cookie_header_from_jar(&Self::cookie_jar_path()) {
+            let cookie_header = match Self::cookie_header_from_jar(&self.cookie_jar) {
                 Ok(header) => header,
                 Err(err) => {
                     runtime::warn(format!("failed to parse Yahoo cookie jar: {err}"));
@@ -284,6 +291,12 @@ fn jitter() -> Duration {
     Duration::from_millis(fastrand::u64(0..100))
 }
 
+impl Drop for YahooClient {
+    fn drop(&mut self) {
+        let _ = std::fs::remove_file(&self.cookie_jar);
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::YahooClient;

crates/idx-core/src/cache.rs

@@ -136,7 +136,14 @@ impl Cache {
         };
         let raw = serde_json::to_string_pretty(&entry)
             .map_err(|e| IdxError::ParseError(e.to_string()))?;
-        fs::write(path, raw).map_err(|e| IdxError::Io(e.to_string()))
+        // Write then rename, so a concurrent reader never sees a partial
+        // entry (and deletes it as corrupt).
+        let temp = path.with_extension(format!("json.{}.tmp", fastrand::u64(..)));
+        fs::write(&temp, raw).map_err(|e| IdxError::Io(e.to_string()))?;
+        fs::rename(&temp, &path).map_err(|e| {
+            let _ = fs::remove_file(&temp);
+            IdxError::Io(e.to_string())
+        })
     }
 
     pub fn info(&self) -> Result<CacheInfo, IdxError> {
@@ -244,10 +251,30 @@ impl Cache {
     }
 
     fn entry_path(&self, data_type: &str, symbol: &str) -> PathBuf {
-        self.root.join(data_type).join(format!("{symbol}.json"))
+        self.root
+            .join(safe_file_name(data_type))
+            .join(format!("{}.json", safe_file_name(symbol)))
     }
 }
 
+/// Keys become single path components: anything that could separate or
+/// climb directories is percent-encoded. Keys made of letters, digits, and
+/// `._-:^=` (every key idx produces) are unchanged.
+fn safe_file_name(key: &str) -> String {
+    let mut out = String::with_capacity(key.len());
+    for (i, byte) in key.bytes().enumerate() {
+        let safe = byte.is_ascii_alphanumeric()
+            || matches!(byte, b'_' | b'-' | b':' | b'^' | b'=')
+            || (byte == b'.' && i > 0);
+        if safe {
+            out.push(byte as char);
+        } else {
+            out.push_str(&format!("%{byte:02X}"));
+        }
+    }
+    out
+}
+
 fn is_expired(fetched_at: DateTime<Utc>, ttl_secs: u64) -> bool {
     let ttl =
         chrono::Duration::from_std(Duration::from_secs(ttl_secs)).unwrap_or(chrono::Duration::MAX);
@@ -275,7 +302,7 @@ mod tests {
 
     use serde::{Deserialize, Serialize};
 
-    use super::{Cache, CacheEntry};
+    use super::{Cache, CacheEntry, safe_file_name};
 
     static TMP_COUNTER: AtomicUsize = AtomicUsize::new(0);
 
@@ -377,4 +404,22 @@ mod tests {
         // The corrupted file should be deleted
         assert!(!path.exists(), "corrupted file should be deleted");
     }
+
+    #[test]
+    fn cache_keys_cannot_leave_the_cache_root() {
+        assert_eq!(safe_file_name("BBCA.JK-3mo-1d"), "BBCA.JK-3mo-1d");
+        assert_eq!(
+            safe_file_name("top-performers:id:50"),
+            "top-performers:id:50"
+        );
+        assert_eq!(safe_file_name("../../REPORT"), "%2E.%2F..%2FREPORT");
+        assert_eq!(safe_file_name(".."), "%2E.");
+
+        let cache = Cache::at(std::path::PathBuf::from("/cache/idx"));
+        let path = cache.entry_path("msn-quote", "../../REPORT");
+        assert_eq!(
+            path.parent().unwrap(),
+            std::path::Path::new("/cache/idx/msn-quote")
+        );
+    }
 }

crates/idx-core/src/ownership/archive.rs

@@ -1,5 +1,5 @@
 use std::fs;
-use std::io::{Cursor, Read};
+use std::io::Cursor;
 use std::path::Path;
 
 use chrono::NaiveDate;
@@ -177,11 +177,7 @@ fn extract_txt_from_zip(path: &Path) -> Result<String, IdxError> {
             continue;
         }
 
-        let mut output = String::new();
-        file.read_to_string(&mut output).map_err(|e| {
-            IdxError::ParseError(format!("failed to decode KSEI archive TXT entry: {e}"))
-        })?;
-        return Ok(output);
+        return super::read_zip_entry_bounded(&mut file, "KSEI archive TXT entry");
     }
 
     Err(IdxError::ParseError(
@@ -201,7 +197,8 @@ fn validate_header(header: &str) -> Result<(), IdxError> {
 
 fn parse_archive_date(raw: &str) -> Result<NaiveDate, IdxError> {
     let trimmed = raw.trim();
-    if trimmed.len() != 11 {
+    // Fixed byte offsets below; non-ASCII input would split a character.
+    if trimmed.len() != 11 || !trimmed.is_ascii() {
         return Err(IdxError::ParseError(format!(
             "invalid KSEI archive date `{trimmed}`"
         )));
@@ -357,4 +354,14 @@ mod tests {
         assert!(pdf_local_cp.total_shares <= archive_local_cp.total_shares);
         assert!(pdf_local_cp.percentage_bps <= archive_local_cp.percentage_bps);
     }
+
+    #[test]
+    fn non_ascii_archive_dates_are_parse_errors_not_panics() {
+        // 11 bytes, but byte offsets would split the 'é'.
+        assert!(super::parse_archive_date("1éJan-2026").is_err());
+        assert_eq!(
+            super::parse_archive_date("27-FEB-2026").unwrap(),
+            chrono::NaiveDate::from_ymd_opt(2026, 2, 27).unwrap()
+        );
+    }
 }

crates/idx-core/src/ownership/entities.rs

@@ -81,7 +81,10 @@ pub fn parse_id_percentage(s: &str) -> Result<i64, IdxError> {
         IdxError::ParseError(format!("failed to parse Indonesian percentage '{s}': {e}"))
     })?;
 
-    Ok((whole_i * 100) + frac_i)
+    whole_i
+        .checked_mul(100)
+        .and_then(|bps| bps.checked_add(frac_i))
+        .ok_or_else(|| IdxError::ParseError(format!("Indonesian percentage '{s}' is out of range")))
 }
 
 /// Parse KSEI date format to NaiveDate.
@@ -453,4 +456,10 @@ mod tests {
 
         assert!(normalize_ksei_row(&raw).is_err());
     }
+
+    #[test]
+    fn huge_percentages_are_parse_errors_not_overflow() {
+        assert!(super::parse_id_percentage("9223372036854775807,99").is_err());
+        assert_eq!(super::parse_id_percentage("54,94").unwrap(), 5494);
+    }
 }

crates/idx-core/src/ownership/mod.rs

@@ -8,3 +8,27 @@ pub mod search;
 pub mod snapshot;
 pub mod types;
 pub mod xlsx;
+
+use std::io::Read;
+
+use crate::error::IdxError;
+
+/// Largest decompressed ZIP entry read into memory. Real IDX worksheets and
+/// KSEI TXT archives are tens of MB; this bounds a hostile "zip bomb".
+const MAX_ZIP_ENTRY_BYTES: u64 = 256 * 1024 * 1024;
+
+/// Reads one decompressed ZIP entry as UTF-8, failing past the size limit.
+pub(crate) fn read_zip_entry_bounded(entry: impl Read, what: &str) -> Result<String, IdxError> {
+    let mut bytes = Vec::new();
+    entry
+        .take(MAX_ZIP_ENTRY_BYTES + 1)
+        .read_to_end(&mut bytes)
+        .map_err(|e| IdxError::ParseError(format!("failed to read {what}: {e}")))?;
+    if bytes.len() as u64 > MAX_ZIP_ENTRY_BYTES {
+        return Err(IdxError::ParseError(format!(
+            "{what} expands beyond {MAX_ZIP_ENTRY_BYTES} bytes"
+        )));
+    }
+    String::from_utf8(bytes)
+        .map_err(|e| IdxError::ParseError(format!("failed to decode {what}: {e}")))
+}

crates/idx-core/src/ownership/snapshot.rs

@@ -202,6 +202,14 @@ pub fn sync_snapshot(
 
     let install_result = (|| -> Result<(), IdxError> {
         validate_snapshot_db(&temp_path, &manifest.snapshot)?;
+        // The decision was made before the download; an import or edit that
+        // landed since must not be overwritten without --force.
+        if !force && !same_local_state(local_state.as_ref(), inspect_local_db(db_path)?.as_ref()) {
+            return Err(IdxError::DatabaseError(
+                "the local ownership database changed during sync; rerun `idx ownership sync`"
+                    .to_string(),
+            ));
+        }
         install_snapshot_file(&temp_path, db_path)
     })();
 
@@ -362,6 +370,21 @@ fn inspect_local_db(db_path: &Path) -> Result<Option<LocalSnapshotState>, IdxErr
     }))
 }
 
+fn same_local_state(
+    before: Option<&LocalSnapshotState>,
+    after: Option<&LocalSnapshotState>,
+) -> bool {
+    match (before, after) {
+        (None, None) => true,
+        (Some(before), Some(after)) => {
+            before.file_sha256.is_some()
+                && before.file_sha256 == after.file_sha256
+                && before.release_count == after.release_count
+        }
+        _ => false,
+    }
+}
+
 /// Moves committed WAL frames into the database file, so its hash covers local
 /// edits too. Returns false when changes remain in the WAL.
 fn checkpoint_wal(db_path: &Path) -> bool {
@@ -446,10 +469,12 @@ fn build_sync_decision(
         };
     }
 
-    if snapshot
-        .replaces_sqlite_sha256
-        .iter()
-        .any(|sha256| file_sha256.is_some_and(|local| sha256.eq_ignore_ascii_case(local)))
+    // A rebuild never silently drops history the local database has.
+    if snapshot.release_count >= local.release_count
+        && snapshot
+            .replaces_sqlite_sha256
+            .iter()
+            .any(|sha256| file_sha256.is_some_and(|local| sha256.eq_ignore_ascii_case(local)))
     {
         return SyncDecision {
             action: OwnershipSyncAction::Updated,
@@ -909,6 +934,20 @@ mod tests {
         assert!(!decision.should_download);
     }
 
+    #[test]
+    fn build_sync_decision_never_drops_history_through_a_rebuild() {
+        // A forced same-date rebuild with fewer releases lists the old build;
+        // a client holding it keeps its history instead of downgrading.
+        let earlier = "e".repeat(64);
+        let mut snapshot = same_date_snapshot(vec![earlier.clone()]);
+        snapshot.release_count = 1;
+        let mut local = same_release_local(&snapshot, &earlier);
+        local.release_count = 3;
+
+        let decision = build_sync_decision(Some(&local), &snapshot, false);
+        assert!(!decision.should_download, "{}", decision.reason);
+    }
+
     #[test]
     fn build_sync_decision_keeps_a_locally_modified_database() {
         // Same release metadata but a file matching no published build (local

crates/idx-core/src/ownership/xlsx.rs

@@ -9,7 +9,7 @@
 
 use std::collections::HashMap;
 use std::fs;
-use std::io::{Cursor, Read};
+use std::io::Cursor;
 use std::path::Path;
 
 use chrono::{Duration, NaiveDate};
@@ -83,11 +83,7 @@ fn read_zip_entry(
             )));
         }
     };
-    let mut output = String::new();
-    file.read_to_string(&mut output).map_err(|e| {
-        IdxError::ParseError(format!("failed to decode ownership XLSX entry {name}: {e}"))
-    })?;
-    Ok(Some(output))
+    super::read_zip_entry_bounded(&mut file, &format!("ownership XLSX entry {name}")).map(Some)
 }
 
 fn first_worksheet_name(zip: &mut zip::ZipArchive<Cursor<&[u8]>>) -> Result<String, IdxError> {