Subagent session filenames collide due to session_id[:8] truncation

[fabianyvidal]

Bug

In extract_claude_logs.py (1.1.2), output filenames are generated as:

filename = f"claude-conversation-{date_str}-{session_id[:8]}.md"

For most sessions session_id is a UUID and [:8] gives a unique hex prefix. But Claude Code's subagent sessions use named IDs like agent-aabb12cd, where [:8] strips the disambiguating tail and yields agent-aa. Different subagents on the same date with the same 8-char prefix collide — the second write silently overwrites the first.

Repro

On my machine (~/.claude/projects/ contains 3009 session JSONLs):

• 1366 of them are subagent sessions (agent-*.jsonl)
• After claude-extract --all --output ~/exports, only ~2030 markdown files remain
• ~979 subagent sessions are silently lost (33% of total history)

Distribution of 8-char prefixes among subagent files:

```
209 agent-ac
87 agent-aa
85 agent-af
82 agent-a8
82 agent-a0
81 agent-a7
79 agent-a5
78 agent-ad
78 agent-a6
76 agent-a9
```

Each row above represents one surviving file out of N originals.

Fix

Use the full `session_id` stem rather than `[:8]`. Optionally append a short hash of the parent project directory to disambiguate sessions from different projects that happen to share an agent name.

I implemented a standalone replacement here — happy to send a PR against this repo if useful.

Impact

For users of subagents (which is most heavy users of Claude Code's Task tool), this means a third or more of their conversation history is missing from exports without any warning or error.