feat: add Cortex Code session support

[andrewwowens]

Summary

Adds complete parser and sync support for Snowflake Cortex Code sessions.

Changes

Backend

• Parser (internal/parser/cortex.go): Complete session parser handling both embedded history and split-file formats
  • Parses messages with tool calls and tool results
  • Filters internal blocks and system reminders
  • Supports backup file detection and skipping
  • Extracts session metadata and timestamps from both .json and .history.jsonl files
• Types (internal/parser/types.go): Added AgentCortex constant and registry entry
• Sync Engine (internal/sync/engine.go): Added processCortex function and wired into agent switch

Frontend

• Agent Definitions (frontend/src/lib/utils/agents.ts): Added Cortex with cyan color and "Cortex Code" label

Documentation

• README.md: Added Cortex Code to supported agents table

Session Discovery

• Default directory: ~/.snowflake/cortex/conversations/
• Environment variable: CORTEX_DIR
• Config key: cortex_dirs
• ID prefix: cortex:

File Format Support

Handles both Cortex session formats:

1. Embedded history: Large .json files with history array
2. Split-file format: Metadata .json + .history.jsonl companion file
3. Backup files: Automatically skips .back.*.json files

Testing

Tested with real Cortex sessions:

• ✅ 12 sessions discovered and synced
• ✅ 2,164 messages parsed
• ✅ 1,288 tool calls extracted and categorized
• ✅ Both file formats working correctly
• ✅ Backup files properly skipped
• ✅ Frontend displaying sessions with correct metadata

Screenshots

Session list showing Cortex sessions with cyan badge alongside Claude and Codex sessions.

[roborev-ci]

roborev: Combined Review (b1eabeb)

Verdict: Changes are not ready to merge; there is one High-severity sync/cache regression in the new Cortex session handling.

High

• Cortex sessions will bypass sync caching and be reparsed on every sync
  • Location: internal/parser/cortex.go#L405, internal/parser/cortex.go#L497, internal/sync/engine.go#L1897
  • Problem: Split Cortex sessions persist ParsedSession.File metadata from <uuid>.history.jsonl, while discovery and sync skip/hash logic still operate on <uuid>.json. That makes the stored path/mtime/size inconsistent with the file the sync engine checks, so shouldSkipByPath will never match for split sessions and they can be reparsed on every sync tick.
  • Fix: Keep ParsedSession.File aligned with the discovered .json file, or update discovery/hash/skip logic so it consistently uses the .history.jsonl companion when present.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (5d1e019)

Verdict: 2 medium-severity issues should be addressed before merge.

Medium

• internal/parser/cortex.go:537 / internal/parser/cortex.go:552: FindCortexSourceFile validates sessionID before stripping the optional cortex: prefix, so IDs stored as cortex:<uuid> fail lookup even though the function appears intended to accept that form. Trim the prefix before calling IsValidSessionID.

• internal/parser/cortex.go:352: When a split session has no embedded history, failures reading the companion .history.jsonl file are silently treated as “session missing” by returning nil, nil, nil. This can hide permission errors or transient I/O failures and cause sessions to disappear from sync without an actionable error. Only suppress os.IsNotExist; propagate other read errors.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)

[roborev-ci]

roborev: Combined Review (df9f864)

Verdict: No medium-or-higher issues found; this change looks clean based on the submitted reviews.

---

Synthesized from 3 reviews (agents: codex, gemini | types: default, security)