Fix Cursor CLI activity trace grouping

[Minitour]

Summary

• Reconcile Cursor CLI activity rows so prompts and tool spans stay in one run during live SSE updates (chat vs agent-session conversation_id mismatch).
• PR Release: v2.0.1 #168 review fixes: valid JSON (or null) for oversized attributes_json; plugin skill fallback only for type: plugin; composite run ids (conversation:generation); real manifestFile in nested plugin lookup; test formatting.
• Skip Cursor conversation reconcile when the same generation_id maps to conflicting chat ids.

Test plan

• bun test src/shared/__tests__/activity-correlation-reconcile.test.ts src/shared/__tests__/activity-attributes.test.ts web-ui/src/features/projects/components/activity/groupActivityRuns.test.ts src/server/__tests__/skill-content-plugin.test.ts
• Cursor Agent CLI: opening prompt remains visible as tools stream in
• Activity run dialog: two conversations with the same generation_id open distinct runs

[qodo-free-for-open-source-projects]

PR Summary by Qodo

Fix Cursor CLI activity trace grouping via conversation ID reconciliation

🐞 Bug fix 🧪 Tests 🕐 20-40 Minutes

AI Description

• Reconcile Cursor activity rows so one generation uses a single chat conversation_id.
• Apply reconciliation before UI grouping to keep SSE tool spans in the same run.
• Add unit tests covering reconciliation and Cursor CLI grouping behavior.

Diagram

graph TD
  cursor{{"Cursor Agent CLI"}} --> rows["Activity hook rows"] --> reconcile["Reconcile conv IDs"] --> group["Group activity runs"] --> ui["Activity feed UI"]
  sharedTests["Shared reconcile tests"] --> reconcile
  uiTests["UI grouping tests"] --> group

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Fix the Cursor CLI hook emitter upstream

• ➕ Eliminates the need for downstream reconciliation logic
• ➕ Prevents inconsistent data from entering any consumers (UI, storage, analytics)
• ➖ May not be actionable/owned by this repo
• ➖ Doesn’t help users until Cursor ships the change

2. Reconcile at ingestion/storage time (server-side)

• ➕ Ensures DB rows are consistent for all clients and historical queries
• ➕ Avoids needing UI-side normalization in multiple places
• ➖ More invasive; may require backfill/migration strategy for existing rows
• ➖ Higher risk if reconciliation logic has edge cases across sources

3. Group primarily by generation_id instead of conversation_id rewriting

• ➕ Avoids mutating/rewriting identifiers
• ➕ Can be localized to the grouping algorithm
• ➖ Requires broader grouping rule changes and careful handling of multi-conversation generations
• ➖ May complicate other sources/providers where generation_id semantics differ

Recommendation: The PR’s approach (provider-scoped reconciliation for Cursor, applied before grouping) is a pragmatic fix with low blast radius and directly addresses the live SSE split-run symptom. Long term, consider moving the same normalization server-side at ingestion so all consumers (and historical data) benefit, but the current UI-level normalization is an appropriate, fast remediation.

Files changed (4) +224 / -1

Bug fix (2) +99 / -1

activity-correlation-reconcile.tsImplement Cursor activity conversation_id reconciliation by generation +96/-0

Implement Cursor activity conversation_id reconciliation by generation

• Adds shared utilities to extract chat conversation IDs from Cursor transcript paths and to normalize Cursor activity rows so a generation’s prompt/tools/stops share the same chat conversation_id. Uses a two-pass mapping (generation -> chat, then agent-session -> chat) to reconcile tool spans even when they follow rewritten rows.

src/shared/activity-correlation-reconcile.ts

groupActivityRuns.tsNormalize Cursor rows before grouping activity conversations +3/-1

Normalize Cursor rows before grouping activity conversations

• Imports and applies the shared Cursor conversation_id reconciliation to the incoming calls before correlating/grouping. This prevents live SSE updates from splitting prompts and tool spans into separate activity runs.

web-ui/src/features/projects/components/activity/groupActivityRuns.ts

Tests (2) +125 / -0

activity-correlation-reconcile.test.tsAdd unit tests for Cursor conversation ID reconciliation +93/-0

Add unit tests for Cursor conversation ID reconciliation

• Introduces tests for transcript-path chat ID parsing and for rewriting Cursor tool/shell rows to the prompt chat conversation_id within a generation. Includes a guard test ensuring non-Cursor sources are not modified.

src/shared/tests/activity-correlation-reconcile.test.ts

groupActivityRuns.test.tsAdd regression test for Cursor CLI prompt/tool span grouping +32/-0

Add regression test for Cursor CLI prompt/tool span grouping

• Adds a test ensuring a Cursor CLI tool span with an agent-session conversation_id merges into the prompt’s chat conversation_id for the same generation, resulting in a single grouped run.

web-ui/src/features/projects/components/activity/groupActivityRuns.test.ts

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0) 🎨 UX issues (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

1. Unneeded JSON parsing ✓ Resolved 🐞 Bug ➹ Performance

Description

reconcileCursorActivityConversationIds() JSON.parses attributes_json for every Cursor row with a
generation_id that isn’t a prompt with a conversation_id, even when transcript_path is absent.
Because groupActivityConversations() runs this reconciliation on every activity grouping, it adds
avoidable CPU overhead during UI rendering.

Code

src/shared/activity-correlation-reconcile.ts[R65-68]

+		if (!chatId) {
+			chatId = chatConversationIdFromTranscriptPath(
+				transcriptPathFromRow(call),
+			);

Evidence

The reconciliation loop calls transcriptPathFromRow(call) whenever a Cursor row isn’t a prompt
with a usable conversation_id; transcriptPathFromRow() unconditionally calls
parseAttributesJson(), which runs JSON.parse() for any non-empty attributes_json. This work is
triggered for the entire activity list because groupActivityConversations() runs reconciliation
before grouping.

web-ui/src/features/projects/components/activity/groupActivityRuns.ts[38-47]
src/shared/activity-correlation-reconcile.ts[26-44]
src/shared/activity-correlation-reconcile.ts[56-69]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
`reconcileCursorActivityConversationIds()` ends up calling `JSON.parse()` on `attributes_json` for many Cursor rows (any non-prompt row with a non-empty `generation_id`) just to attempt to extract `transcript_path`. In typical feeds, most rows won’t contribute a `transcript_path`, so this adds avoidable CPU work in the browser each time `groupActivityConversations()` is called.
### Issue Context
- In `groupActivityConversations()`, reconciliation runs before grouping.
- In reconciliation, `transcriptPathFromRow()` always calls `parseAttributesJson()` which calls `JSON.parse()` when `attributes_json` is non-empty.
### Fix Focus Areas
- src/shared/activity-correlation-reconcile.ts[26-44]
- src/shared/activity-correlation-reconcile.ts[56-70]
### Suggested approach
- Only attempt transcript-path extraction on rows that are likely to carry it (e.g. `kind === "stop"`), **or**
- Add a cheap pre-check before parsing (e.g. `attributes_json.includes('"transcript_path"')`) to avoid `JSON.parse()` when the key can’t be present.
- Keep behavior unchanged for correctness (still learn chat id from prompt rows and/or transcript_path when available).

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

_{To customize comments, go to the Qodo configuration screen, or learn more in the docs.}