fix(coding-agent): prevent duplicate session prefixes and /tree freezes

[w-winter]

Problem

/tree freezes the TUI permanently (no input or paint until the tab is closed) on some sessions. The affected sessions' .jsonl files contain duplicated session headers and duplicate entry IDs. /tree does synchronous traversals that assume a well-formed tree, so duplicates and cycles cause non-terminating loops on the UI thread.

How duplicated headers happen

Extensions can call pi.appendEntry(...) before the first assistant message. For example, examples/extensions/preset.ts persists state on turn_start, which fires before the assistant responds. Session persistence is deferred until the first assistant message. When that flush fires, _persist() appends the full in-memory fileEntries to disk, but the file already has content from those early appendEntry calls. Result: duplicated session header, prefix entries, and entry IDs.

The added test "rewrites deferred pre-assistant state instead of duplicating the existing session prefix" reproduces this directly.

Fix

This prevents the duplication and hardens traversal so that existing malformed sessions don't hang.

Persistence: first deferred flush now rewrites the file instead of appending, so early extension entries are preserved without duplication.

Traversal guards:

• buildSessionContext() and getBranch(): visited-ID set terminates leaf-to-root walks on cycles
• getTree(): dedupes entries by ID (last occurrence wins), breaks parent cycles into roots, memoizes cycle detection for large sessions

Validation

• npm run check
• cd packages/coding-agent && npx vitest run test/session-manager/file-operations.test.ts
• cd packages/coding-agent && npx vitest run test/session-manager/build-context.test.ts
• cd packages/coding-agent && npx vitest run test/session-manager/tree-traversal.test.ts

[github-actions]

Hi @w-winter, thanks for your interest in contributing!

We ask new contributors to open an issue first before submitting a PR. This helps us discuss the approach and avoid wasted effort.

Next steps:

1. Open an issue describing what you want to change and why (keep it concise, write in your human voice, AI slop will be closed)
2. Once a maintainer approves with lgtm, you'll be added to the approved contributors list
3. Then you can submit your PR

This PR will be closed automatically. See https://github.com/badlogic/pi-mono/blob/main/CONTRIBUTING.md for more details.

[badlogic]

This papers over the actual issue: somehow a corrupted session was created. I need to understand how that happened. This is not the fix we are looking for. Can report on #contributors on discord. pinged you there.

[badlogic]

Thanks @w-winter for the report and investigation. The root cause is right: createBranchedSession() sets flushed = true on the new file, then _persist()'s no-assistant guard resets it to false, and the subsequent flush appends all in-memory entries to the already-populated file.

We went with a narrower fix: createBranchedSession() now defers file creation when the branched path has no assistant message, matching the newSession() contract. This avoids the flushed state mismatch entirely rather than changing the flush strategy in _persist().

The traversal guards (cycle detection in getTree/getBranch/buildSessionContext) aren't needed since the root cause is fixed and we don't produce malformed sessions. Closing in favor of 2f64df1.