[rudycelekli]

What type of PR is this? (check all applicable)

• Refactor
• Feature
• Bug Fix
• Optimization
• Documentation Update

Related issue: N/A (found by reading the code; reproduced on main at 6b33dff)

What changed?
Treat $patch records like $rewindTo: append them to journal_controls and do not apply them, consistent with the module's stated policy of retaining recorded activity across rewinds (masking/removing model context does not undo actions). Tool results still come from full message upserts (recordToolCalls -> pushMessage), which are unchanged.

Why?
Gemini-cli (google-gemini/gemini-cli#29568, merged 2026-10-01, packages/core/src/services/chatRecordingTypes.ts MessagePatchRecord, chatRecordingService.ts:1587 appendRecord({ $patch: patchPayload })) now appends {"$patch": {updates, removeIds, orderIds}} records when it syncs history (tool-output masking, removed/reordered turns). GeminiParser.parse_file (Sensor/adr_sensor/parsers/gemini_parser.py:143-156) only special-cases $rewindTo and $set; any other record without "id" is treated as a metadata update, so the whole patch lands in session_context.session_metadata["$patch"], each patch overwriting the previous one, and it is not listed in journal_controls.

How did you test it?

• Reproduction: new tests/test_gemini_parser.py::test_history_patch_is_a_journal_control_not_session_metadata. main: FAILED ('$patch' not in {...session_metadata...}); branch: passed (patch in journal_controls, original tool result and both messages kept).
• Test suite: Sensor pytest tests/ -q (py3.14) 609 passed; ruff clean; Python 3.9.6 parser test files 245 passed.
• Not tested: a real gemini-cli journal with $patch (no Gemini CLI data locally); record shape taken from the upstream source above. Judgement call for the lead: an alternative would be to apply content/toolCalls patches, but that would replace recorded tool output with masked placeholders, against the parser's documented retention policy.

Potential risks
The change only affects the input described above; other paths go through the same code as before, and the component's existing tests pass.

AI assistance: Claude Code (model Claude Opus 5.5) found this bug, wrote the fix and test, and drafted this description; submitted by @rudycelekli.

🤖 Generated with Claude Code

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}