feat: add cortex-code transcript provider

[jamescha-earley]

Summary

• Adds a new cortexCode transcript provider that discovers and parses Snowflake Cortex Code conversation transcripts from ~/.snowflake/cortex/conversations/*.json
• Scopes discovery to workspace via working_directory matching, detects running/idle/completed status from last_updated recency and active tool calls
• Extracts connection name, message count, tool call count, and session type as metadata; registers as a default provider alongside cursor, claude-code, codex, and opencode

Test plan

• 41 new tests across 4 test files (schemas, discovery, transcripts, provider integration)
• npm run check passes (biome + typecheck + 262 tests across 36 files)
• Build succeeds with ESM, CJS, and DTS outputs including new providers/cortex-code entry point
• Tested locally against 297 real Cortex Code conversation files — correct discovery, parsing, metadata extraction, and status detection

Checklist

• Tests added
• npm run check passes
• Docs updated (README.md — providers list, entry points, provider count)
• CHANGELOG updated (Unreleased section)

.... Generated with Cortex Code

---

Summary by cubic

Adds a new cortexCode transcript provider that discovers and parses Snowflake Cortex Code session JSONs and normalizes them into agent snapshots. It’s included in the default provider set for observers.

• New Features

  • Discovers sessions from ~/.snowflake/cortex/conversations/*.json, scoped by working_directory.
  • Detects running/idle/completed using last_updated recency and active tool calls.
  • Extracts metadata: connection name, message count, tool call count, session type; builds task summary from first user prompt.
  • Caches file headers by mtime to skip unchanged files; JSON file watch with 150ms debounce.
  • Public entry: @agentprobe/core/providers/cortex-code; registered as a default provider; 41 tests added.

• Migration

  • No changes required—createObserver() now includes cortexCode() by default.
  • To customize, import cortexCode from @agentprobe/core/providers/cortex-code and pass options like cortexHomePath, maxFiles, or disable watch.

^{Written for commit c645adf. Summary will update on new commits.}

[cubic-dev-ai]

1 issue found across 18 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/providers/cortex-code/transcripts.ts">

<violation number="1" location="src/providers/cortex-code/transcripts.ts:237">
P2: Hardcoded source string diverges from the canonical CORTEX_CODE_SOURCE_KIND, risking identifier drift and downstream mismatches.</violation>
</file>

---

Since this is your first cubic review, here's how it works:

• cubic automatically reviews your code and comments on bugs and improvements
• Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
• Add one-off context when rerunning by tagging @cubic-dev-ai with guidance or docs links (including llms.txt)
• Ask questions if you need clarification on any suggestion

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[vtemian]

Thanks for this contribution, James! The provider is architecturally sound: it follows the discover -> read -> normalize pipeline correctly, uses factory functions, accumulates warnings non-fatally, and mirrors the existing claude-code provider patterns well. The test suite is solid (41 tests, real filesystem, proper cleanup). Clearly you studied the codebase before writing this.

A few things need fixing before we can merge:

1. Module-level mutable state — headerCache in discovery.ts is a module-level Map singleton. This leaks across provider instances and test runs. The claude-code provider keeps its cache in closed-over state inside the factory function. Please move headerCache into the factory's closure or pass it through options.

2. Unused export isToolResultError — exported from schemas.ts and re-exported through index.ts, but never called anywhere. YAGNI: please remove it.

3. Magic strings — "Chat for session:" in transcripts.ts and "cortex-code://sessions" in provider.ts should be named constants (per project rules, no magic strings).

4. Unused parameter _fileUpdatedAt — in recalculateStatus in transcripts.ts. If it's not needed, remove it from the signature rather than underscore-prefixing.

5. Version bump — the PR bumps package.json to 0.2.0. Please revert this: version bumps are the maintainer's call.

None of these require structural changes, just a cleanup pass. Looking forward to v2!