feat: parse GitHub.copilot-chat/transcripts/*.jsonl event-stream format

[hora7ce]

Summary

Closes #64.

VS Code stores Copilot Chat sessions in two distinct locations inside each workspace's workspaceStorage entry:

#	Location	Format	Status
1	chatSessions/*.{json,jsonl}	JSON / JSONL session blobs	✅ Already parsed
2	GitHub.copilot-chat/transcripts/*.jsonl	Newline-delimited event stream	❌ Silently ignored — this PR

Sessions recorded only in format 2 never appeared in the dashboard. This PR makes the extension parse both formats transparently.

---

The transcript event-stream format

Each .jsonl file represents one session. Each line is a typed JSON event:

Event type	Meaning
session.start	New conversation; carries sessionId and metadata
user.message	User turn with content text
assistant.message	AI response; may include toolRequests[] array
tool.execution_start	Tool call begins; carries toolCallId + toolName
tool.execution_complete	Tool call finishes

---

Changes

src/core/parser-vscode.ts

New private helpers — each kept small to stay within the ESLint complexity limit:

Helper	Purpose
listTranscriptFiles(dir)	Lists *.jsonl files under a transcripts/ directory; returns [] when the dir does not exist
parseTranscriptLines(raw)	Deserialises the raw JSONL text; blank and corrupt lines are silently skipped
buildToolNameIndex(events)	Pre-indexes toolCallId → toolName from tool.execution_start events
collectToolsFromToolRequests(...)	Extracts tool names from assistant.message.toolRequests[], with toolCallId fallback to the pre-built index
buildRequestsFromTranscriptEvents(events, toolNames)	Groups events into SessionRequest[]; each user.message starts a new turn that is flushed when the next user.message (or end-of-stream) arrives

New exported function:

• parseTranscriptFile(filePath, wsId, wsName, harness, customInstructionsBytes?) — reads a transcript file, builds the Session object (or returns null for empty / unreadable files), and is the single public surface consumed by the integration points below.

Integration in processWorkspaceEntry and processWorkspaceEntryAsync:

• After processing chatSessions/, both functions now also scan GitHub.copilot-chat/transcripts/ and add discovered sessions to the same sessions[] / sessionSourceIndex pipeline — fully transparent to the rest of the analyzer.
• The async path accounts for transcript files in totalUnits so progress reporting stays accurate.

src/core/parser-vscode.test.ts

Five new tests in a parseTranscriptFile describe block:

Test	Acceptance criterion covered
Full flow — single turn with tool call	Maps event-stream turns to SessionData.requests[]; validates sessionId, workspaceId, harness, messageText, responseText, toolsUsed, agentMode, timestamp
Multi-turn grouping	Two user/assistant pairs produce two separate requests
No user messages	Returns null gracefully
All-corrupt JSONL lines	Returns null gracefully (guards the events.length === 0 path)
Tool deduplication	Same tool name appearing in both toolRequests and tool.execution_start collapses to a single entry

---

Dependency

This branch is based on PR #63 (fix/vscode-server-log-discovery), which adds ~/.vscode-server path discovery. The two PRs are independent at the code level — they touch disjoint functions — but merging #63 first is recommended so the transcript sessions are attributed the correct Local Agent (Server) harness label for VS Code Server users.

---

Checklist

• npm run check passes (typecheck + lint + spellcheck + knip + tests)
• Changes are covered by tests (5 new cases; all acceptance criteria from feat: parse GitHub.copilot-chat/transcripts/*.jsonl event-stream format #64 met)
• Documentation updated (n/a — no new harness names or user-visible doc changes introduced by this PR)

[aymenfurter]

@san360 LGTM can you check if this PR works and does not introduce any duplication? Thanks!

[eyehiel]

I took this PR since my data was empty and with this PR it parses my data,
but from some reason I see duplicated sessions in the Timeline:

[san360]

Thanks @hora7ce for putting this together, and @eyehiel for the duplicate-sessions screenshot — both were extremely useful.

After reviewing the PR locally on Windows desktop VS Code (built from this branch and from main for comparison), the parser does work and surfaces sessions that were previously invisible. However, in its current form it pushes transcript sessions into the same sessions[] array as the existing chatSessions/ parser without deduplicating by sessionId, so every conversation that exists in both formats now appears twice in the Timeline. Reproduced locally; matches @eyehiel's report.

Before landing the transcript parser, we need to answer a set of open questions about when each format is present, whether the same sessionId is shared across them, and what the correct merge/dedup strategy should be (single-source, dual-source with dedup, or dual-source with enrichment-merge).

I've opened #87 to track that investigation in detail, including:

• background on why VS Code writes both formats (Format 2 = Agent Debug Logs stream introduced in VS Code 1.110 / Copilot Chat 0.38, Feb 2026),
• the specific edge cases to verify (CLI sessions, pruned chatSessions/, Remote/WSL global path, Insiders, etc.),
• a proposed direction,
• a definition of done.

Closing this PR in favor of #87 — the code here is a good starting point and will be referenced from the issue. Please feel free to follow up on #87 with your environment details (it sounds like you're on WSL / Remote-SSH per #62 / #63, which is exactly the scenario we most need data from). Also note one test (findVsCodeDirs > includes server workspaceStorage paths on non-Windows hosts) needs a skipIf(process.platform === 'win32') guard or a rework — that should be folded into the follow-up.

Closing as: more investigation required (tracked in #87).