emrg: revert tool result sliding window + DeepSeek reasoning pass-back (#936 #937)

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.

[argszero]

Revert #936 (commit 6d4f3bc) and #937 (commit e6c066b) — restore v0.2.67 behavior (host rant 2026-08-22T21:23:18).

Reason: #936 (tool result sliding window) and #937 (unrestricted reasoning_content pass-back) together cause 413 request-too-large on long sessions (3 assistant reasoning_content blocks ≈ 2.54MB ≈ 75% of a 3.39MB payload) and multi-round tool-task anomalies. Host decision: revert both together for the safest restore, prioritizing working new installations.

Scope:

1. Revert emrg: implement tool result sliding window #936: removed _apply_tool_window call (daemon.py), _apply_tool_window/_fold_tool_group methods, tool_window_rounds config (config.py + daemon.py), system.j2 fold-notice block, window tests (test_daemon.py)
2. Revert emrg: persist and pass back DeepSeek thinking-mode reasoning #937: session.py reasoning_content pass-back removed; daemon.py 3 append_message sites no longer persist reasoning; in-memory reasoning_content pass-back removed; emrg: persist and pass back DeepSeek thinking-mode reasoning #937 tests removed
3. Kept: emrg: llm.jsonl response records capture reasoning (think block) + usage.reasoning_tokens #833 reasoning accumulation (llm.jsonl + logs only, no pass-back, no history persistence)
4. History compat: existing reasoning / fold-placeholder fields naturally ignored

Verification: pytest 1002 passed + 1 skipped (1003 collected, matches Agent.md); import/CLI OK. CI: test + test-windows re-running on updated head.

[argszero]

✅ LGTM — cycle 53. Verified: exact inverse of #937 (4 files, 4 insertions / 83 deletions): session.py reasoning_content mapping removed; daemon.py append_message reasoning persistence + in-memory reasoning_content pass-back removed (tool loop, _reflect, _consolidate_session_memories); llm.py reasoning accumulation retained (#833, llm.jsonl/logs only — verified reasoning_parts → response record path intact); #937's 3 tests removed; Agent.md count restored to 1008 matching actual. Local pytest 1007 passed + 1 skipped. CI test + test-windows PASS (run 32575229361).

[argszero]

✅ LGTM — cycle 54. Verified the FULL combined diff (2 commits, 7 files, +5/-328):

• #937 revert (0cc46f4): session.py reasoning_content pass-back removed; daemon.py 3 append_message sites + in-memory reasoning_content pass-back removed; #937 tests removed
• #936 revert (3cd02a9): _apply_tool_window call + methods removed (daemon.py), tool_window_rounds config removed (config.py + daemon.py), system.j2 fold-notice removed, window tests removed (test_daemon.py)
• Zero residual: no reasoning_content in session.py, no tool_window/_fold_tool in config/daemon/system.j2
• #833 reasoning accumulation retained (llm.py reasoning_parts → response record, logs only)
• Agent.md 1003 matches actual pytest count (1002 passed + 1 skipped)
• CI test + test-windows PASS (run 32575814490)

[argszero]

✅ LGTM — cycle 55. Third consecutive approval from a distinct cycle (53/54/55). Fresh verification of full head (2 commits): 7 files, +5/-328 — #937 revert (session.py reasoning_content pass-back removed, daemon.py 3 append_message sites + in-memory pass-back removed) + #936 revert (_apply_tool_window/_fold_tool_group/tool_window_rounds/system.j2 fold-notice/window tests removed). Zero residual (session.py 0 reasoning_content, daemon.py 0 tool_window). #833 reasoning accumulation retained (llm.py, logs only). CI test + test-windows PASS (run 32575814490). MERGEABLE / CLEAN — merging.