From 1313c2a221df8ceba76db48dc913acb183e3ec38 Mon Sep 17 00:00:00 2001
From: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
Date: Mon, 14 Sep 2026 12:46:01 -0700
Subject: [PATCH 1/4] feat: retain current request requirements through
 compaction

Port the semantic request-retention delta from reviewed #2 candidate 42e3cfffdfa0f40e6a27f7fd6fc58979470af21e (implementation commits 482ea0326a33b3a3fcdee9cb0e3514a707cbd032, db63ff2abfc490ab763bf6ca0263d7d40120d347, and 95de3a4d1580f506fbf6669f24520837faa7a7f9).

Original contribution credit: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>

Generated with Amplifier

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
---
 amplifier_module_context_simple/__init__.py | 155 ++++++++-
 tests/test_request_retention.py             | 330 ++++++++++++++++++++
 2 files changed, 469 insertions(+), 16 deletions(-)
 create mode 100644 tests/test_request_retention.py

diff --git a/amplifier_module_context_simple/__init__.py b/amplifier_module_context_simple/__init__.py
index 4435d13..498bcbf 100644
--- a/amplifier_module_context_simple/__init__.py
+++ b/amplifier_module_context_simple/__init__.py
@@ -46,6 +46,7 @@
 from typing import Any
 
 from amplifier_core import ModuleCoordinator, TextBlock
+from amplifier_core.llm_errors import ContextLengthError
 
 logger = logging.getLogger(__name__)
 
@@ -107,6 +108,16 @@ def _carries_loaded_tool_state(msg: dict[str, Any]) -> bool:
     return any(meta.get(key) for key in LOADED_TOOL_STATE_METADATA_KEYS)
 
 
+def _is_human_message(msg: dict[str, Any]) -> bool:
+    """Wire role alone does not distinguish a prompt from an injection."""
+    meta = msg.get("metadata") or {}
+    return (
+        msg.get("role") == "user"
+        and not meta.get("ephemeral")
+        and meta.get("source") not in ("hook", "context-compaction")
+    )
+
+
 async def mount(coordinator: ModuleCoordinator, config: dict[str, Any] | None = None):
     """
     Mount the simple context manager.
@@ -190,6 +201,12 @@ async def mount(coordinator: ModuleCoordinator, config: dict[str, Any] | None =
         )
 
     await coordinator.mount("context", context)
+    # Optional module capability; the core Context protocol is unchanged.
+    register_capability = getattr(coordinator, "register_capability", None)
+    if callable(register_capability):
+        register_capability(
+            "context.request_retention", context.get_messages_for_request_retaining
+        )
     logger.info(f"Mounted SimpleContextManager (token_meter={token_meter!r})")
 
     async def cleanup() -> None:
@@ -224,14 +241,14 @@ class SimpleContextManager:
     Level 5: Remove more messages (60% of configured protection)
     Level 6: Truncate remaining tool results (except last N)
     Level 7: Remove more messages (30% of configured protection - last resort)
-    Level 8: Stub first user message + remove old stubs (extreme pressure)
+    Level 8: Stub unprotected machine prefix + remove old stubs (extreme pressure)
 
     This interleaved approach ensures minimal data loss by:
     - Preferring truncation (preserves structure) over removal (loses context)
     - Progressively relaxing protection as pressure increases
     - Respecting configured protected_recent as baseline, only relaxing under pressure
-    - Always protecting: system messages, last user message, last N tool results, tool pairs
-    - First user message: stubbable at Level 8, but never fully removed
+    - Always protecting: system messages, first/last human prompts, last N tool results, tool pairs
+    - Requested active persisted injections remain complete through all levels
     """
 
     def __init__(
@@ -328,6 +345,8 @@ def __init__(
         self._last_measured_prompt_tokens: int | None = None
         self._last_token_meter_stats: dict[str, Any] | None = None
         self._system_prompt_factory: Callable[[], Awaitable[str]] | None = None
+        self._request_retained_contents: frozenset[str] = frozenset()
+        self._request_protected_seqs: set[int] = set()
 
         # --- Sticky compaction decision state ---
         # Compaction decisions (remove / truncate / stub) are keyed by a
@@ -553,6 +572,82 @@ async def set_system_prompt_factory(
         self._system_prompt_factory = factory
         logger.info("System prompt factory registered - will refresh on each request")
 
+    async def get_messages_for_request_retaining(
+        self,
+        *,
+        retain_contents: list[str],
+        provider: Any | None = None,
+        token_budget: int | None = None,
+    ) -> list[dict[str, Any]]:
+        """Optional capability: retain current persisted injections for one view.
+
+        Contents must exactly match admitted user-role messages marked
+        ephemeral=True and persisted=True. Only the newest matching copy is
+        protected. This neither admits messages nor changes their lifetime;
+        the caller supplies the current delivery requirements on every call.
+        """
+        previous_contents = self._request_retained_contents
+        previous_seqs = self._request_protected_seqs
+        decisions = (
+            self._removed_seqs.copy(),
+            self._truncated_seqs.copy(),
+            self._stubbed_seqs.copy(),
+            self._sticky_level,
+            self._last_compaction_stats,
+            self._last_token_meter_stats,
+        )
+        try:
+            self._request_retained_contents = frozenset(retain_contents)
+            return await self.get_messages_for_request(token_budget, provider)
+        except BaseException:
+            # A failed/cancelled assembly must not commit a reduction that was
+            # never delivered. Canonical history is unchanged throughout.
+            (
+                self._removed_seqs,
+                self._truncated_seqs,
+                self._stubbed_seqs,
+                self._sticky_level,
+                self._last_compaction_stats,
+                self._last_token_meter_stats,
+            ) = decisions
+            raise
+        finally:
+            self._request_retained_contents = previous_contents
+            self._request_protected_seqs = previous_seqs
+
+    def _protected_sequences(self, messages: list[dict[str, Any]]) -> set[int]:
+        humans = [msg for msg in messages if _is_human_message(msg)]
+        protected = [humans[0], humans[-1]] if humans else []
+        remaining = set(self._request_retained_contents)
+        for msg in reversed(messages):
+            meta = msg.get("metadata") or {}
+            content = msg.get("content")
+            if (
+                msg.get("role") == "user"
+                and meta.get("ephemeral") is True
+                and meta.get("persisted") is True
+                and isinstance(content, str)
+                and content in remaining
+            ):
+                protected.append(msg)
+                remaining.remove(content)
+        if remaining:
+            raise ValueError("Requested retained injection is not in admitted history")
+        return {seq for msg in protected if (seq := self._extract_seq(msg)) is not None}
+
+    def _is_request_protected(self, msg: dict[str, Any]) -> bool:
+        return self._extract_seq(msg) in self._request_protected_seqs
+
+    def _check_retained_budget(
+        self, messages: list[dict[str, Any]], budget: int
+    ) -> None:
+        if self._request_retained_contents and self._estimate_tokens(messages) > budget:
+            raise ContextLengthError(
+                "Context cannot fit the current injections and protected conversation "
+                "within the estimated input budget; shorten the active instructions "
+                "or use a larger context window. Required content was not discarded."
+            )
+
     async def get_messages_for_request(
         self,
         token_budget: int | None = None,
@@ -629,6 +724,16 @@ async def get_messages_for_request(
             # Static mode: use messages as-is (may include stored system messages)
             working_messages = list(self.messages)
 
+        self._request_protected_seqs = self._protected_sequences(working_messages)
+        self._check_retained_budget(
+            [
+                m
+                for m in working_messages
+                if m.get("role") == "system" or self._is_request_protected(m)
+            ],
+            effective_budget,
+        )
+
         token_count, meter_source, estimated_tokens = self._measure_working_tokens(
             working_messages
         )
@@ -643,7 +748,11 @@ async def get_messages_for_request(
         }
 
         # Check if compaction needed (using effective budget with notice reserve deducted)
-        if self._should_compact(token_count, effective_budget):
+        retained_view_over_budget = (
+            bool(self._request_retained_contents)
+            and estimated_tokens > effective_budget
+        )
+        if self._should_compact(token_count, effective_budget) or retained_view_over_budget:
             # Compact EPHEMERALLY - returns new list, working_messages unchanged
             compacted = await self._compact_ephemeral(
                 effective_budget, working_messages
@@ -731,8 +840,10 @@ async def get_messages_for_request(
             # Strip internal bookkeeping at the module boundary -- everything
             # above this point (sticky decisions, token accounting) still runs
             # on messages carrying `_seq`; only what leaves has it removed.
+            self._check_retained_budget(compacted, budget)
             return self._strip_internal_metadata(compacted)
 
+        self._check_retained_budget(working_messages, budget)
         return self._strip_internal_metadata(working_messages)
 
     # Metadata keys that are internal bookkeeping only and must never cross
@@ -876,6 +987,8 @@ def _exceeds_threshold(self, estimated_tokens: int, budget: int) -> bool:
         """
         if budget <= 0:
             return False
+        if self._request_retained_contents and estimated_tokens > budget:
+            return True
         if (
             self.token_meter == TOKEN_METER_ACTUAL
             and self._last_measured_prompt_tokens is not None
@@ -1037,6 +1150,9 @@ def _apply_sticky_decisions(
         result: list[dict[str, Any]] = []
         for msg in messages:
             seq = self._extract_seq(msg)
+            if self._is_request_protected(msg):
+                result.append(dict(msg))
+                continue
             if seq is not None and seq in self._removed_seqs:
                 continue
             if seq is not None and seq in self._truncated_seqs:
@@ -1079,6 +1195,7 @@ async def _compact_ephemeral(
         messages_to_compact = (
             source_messages if source_messages is not None else self.messages
         )
+        self._request_protected_seqs = self._protected_sequences(messages_to_compact)
         target_tokens = int(budget * self.target_usage)
         old_count = len(messages_to_compact)
         old_tokens = self._estimate_tokens(messages_to_compact)
@@ -1420,10 +1537,10 @@ async def _compact_ephemeral(
 
         # Check if we still need more space
         if current_tokens > target_tokens:
-            # === LEVEL 8: Stub first user message + remove old stubs (extreme pressure) ===
+            # === LEVEL 8: Stub unprotected machine prefix + remove old stubs (extreme pressure) ===
             max_level_reached = 8
 
-            # Find first user message and stub it if not already stubbed
+            # Find first and last user messages.
             first_user_idx = None
             last_user_idx = None
             for i, msg in enumerate(working_messages):
@@ -1432,9 +1549,14 @@ async def _compact_ephemeral(
                         first_user_idx = i
                     last_user_idx = i
 
-            # Stub first user message (previously protected) - but NEVER if it's also the last
-            # The last user message is the current intent and must always be preserved
-            if first_user_idx is not None and first_user_idx != last_user_idx:
+            # An unprotected machine prefix can still be reduced at extreme pressure.
+            # Human boundary prompts and current retained injections stay complete.
+            if (
+                first_user_idx is not None
+                and first_user_idx != last_user_idx
+                and not self._is_request_protected(working_messages[first_user_idx])
+                and not _is_human_message(working_messages[first_user_idx])
+            ):
                 first_msg = working_messages[first_user_idx]
                 if not first_msg.get("_stubbed"):
                     content = first_msg.get("content", "")
@@ -1462,6 +1584,7 @@ async def _compact_ephemeral(
                     if msg.get("_stubbed")
                     and i < protected_boundary  # Outside protected recent zone
                     and i != last_user_idx  # Never remove last user message
+                    and not self._is_request_protected(msg)
                 ]
 
                 stubs_removed = 0
@@ -1625,11 +1748,11 @@ def _remove_messages_with_protection(
             i for i, msg in enumerate(messages) if msg.get("role") == "user"
         }
 
-        # Find first and last user message indices (always fully protected from stubbing too)
+        # Human boundaries must not be displaced by machine user-role messages.
         first_user_idx = None
         last_user_idx = None
         for i, msg in enumerate(messages):
-            if msg.get("role") == "user":
+            if _is_human_message(msg):
                 if first_user_idx is None:
                     first_user_idx = i
                 last_user_idx = i
@@ -1654,11 +1777,11 @@ def _remove_messages_with_protection(
                 len(loaded_tool_state_indices),
             )
 
-        # First user message is stubbable at extreme pressure (Level 8), but never fully removed
-        # (It's excluded from removal_candidates via user_message_indices, but can be stubbed)
-        # We don't add it to protected_indices so it can be stubbed at Level 8
-
-        # Always protect the LAST user message (current context)
+        protected_indices.update(
+            i for i, msg in enumerate(messages) if self._is_request_protected(msg)
+        )
+        if first_user_idx is not None:
+            protected_indices.add(first_user_idx)
         if last_user_idx is not None:
             protected_indices.add(last_user_idx)
 
diff --git a/tests/test_request_retention.py b/tests/test_request_retention.py
new file mode 100644
index 0000000..8a41413
--- /dev/null
+++ b/tests/test_request_retention.py
@@ -0,0 +1,330 @@
+"""Required request content must survive actual compaction, not just storage."""
+
+import copy
+
+import pytest
+from amplifier_core.llm_errors import ContextLengthError
+
+from amplifier_module_context_simple import SimpleContextManager, mount
+
+
+def reminder(content):
+    return {
+        "role": "user",
+        "content": content,
+        "metadata": {"ephemeral": True, "persisted": True},
+    }
+
+
+def human(content):
+    return {
+        "role": "user",
+        "content": content,
+        "metadata": {"source": "human"},
+    }
+
+
+def public_envelope(message):
+    """The request view strips only the manager's internal sequence number."""
+    return {
+        **message,
+        "metadata": {
+            key: value
+            for key, value in message["metadata"].items()
+            if key != "_seq"
+        },
+    }
+
+
+async def pressured_context():
+    context = SimpleContextManager(max_tokens=1800, compaction_notice_enabled=False)
+    await context.add_message({"role": "user", "content": "Original task " + "a" * 150})
+    body = "<system-reminders>" + "policy " * 180 + " REQUIRED_FACT</system-reminders>"
+    await context.add_message(reminder(body))
+    for i in range(40):
+        await context.add_message(
+            {"role": "assistant", "content": f"Output {i} " + "x" * 1200}
+        )
+    await context.add_message(
+        {"role": "user", "content": "Current correction " + "b" * 150}
+    )
+    return context, body
+
+
+@pytest.mark.asyncio
+async def test_retention_restores_a_sticky_stub_without_changing_history():
+    context, body = await pressured_context()
+    canonical = copy.deepcopy(await context.get_messages())
+    ordinary = await context.get_messages_for_request()
+    assert not any(m["content"] == body for m in ordinary)
+
+    retained = await context.get_messages_for_request_retaining(retain_contents=[body])
+    assert sum(m["content"] == body for m in retained) == 1
+    assert context._estimate_tokens(retained) <= context.max_tokens
+    assert await context.get_messages() == canonical
+    assert (
+        await context.get_messages_for_request_retaining(retain_contents=[body])
+        == retained
+    )
+
+    # Retention is per request, not a permanent pin that outlives its producer.
+    released = await context.get_messages_for_request_retaining(retain_contents=[])
+    assert not any(m["content"] == body for m in released)
+
+
+@pytest.mark.asyncio
+async def test_machine_messages_cannot_take_human_boundary_protection():
+    context = SimpleContextManager(max_tokens=1400, compaction_notice_enabled=False)
+    first = "Original human request " + "details " * 60 + "ORIGINAL_FACT"
+    last = "Latest human correction " + "details " * 60 + "CURRENT_FACT"
+    await context.add_message(reminder("Earlier machine state " + "m" * 1200))
+    await context.add_message({"role": "user", "content": first})
+    for i in range(35):
+        await context.add_message(
+            {"role": "assistant", "content": f"Work {i} " + "x" * 1000}
+        )
+    await context.add_message({"role": "user", "content": last})
+    await context.add_message(reminder("Later machine state " + "m" * 1200))
+    view = await context.get_messages_for_request()
+    assert any(m["content"] == first for m in view)
+    assert any(m["content"] == last for m in view)
+    assert await context.get_messages_for_request() == view
+
+
+@pytest.mark.asyncio
+async def test_only_newest_matching_copy_is_retained():
+    context, body = await pressured_context()
+    await context.add_message(reminder(body))
+    view = await context.get_messages_for_request_retaining(retain_contents=[body])
+    assert sum(m["content"] == body for m in view) == 1
+    assert sum(m["content"] == body for m in await context.get_messages()) == 2
+
+
+@pytest.mark.asyncio
+async def test_retention_survives_transcript_round_trip():
+    context, body = await pressured_context()
+    transcript = await context.get_messages()
+    resumed = SimpleContextManager(max_tokens=1800, compaction_notice_enabled=False)
+    await resumed.set_messages(transcript)
+    view = await resumed.get_messages_for_request_retaining(retain_contents=[body])
+    assert any(m["content"] == body for m in view)
+
+
+@pytest.mark.asyncio
+async def test_missing_content_fails_without_leaking_a_retention_requirement():
+    context, body = await pressured_context()
+    with pytest.raises(ValueError, match="not in admitted history"):
+        await context.get_messages_for_request_retaining(
+            retain_contents=["not admitted"]
+        )
+    assert not any(
+        m["content"] == body for m in await context.get_messages_for_request()
+    )
+
+
+@pytest.mark.asyncio
+async def test_impossible_retention_fails_before_returning_an_overfull_view():
+    context, body = await pressured_context()
+    huge = "<system-reminders>" + "x" * 12000 + "</system-reminders>"
+    await context.add_message(reminder(huge))
+    with pytest.raises(ContextLengthError, match="Required content was not discarded"):
+        await context.get_messages_for_request_retaining(retain_contents=[huge])
+    assert any(m["content"] == huge for m in await context.get_messages())
+    assert not context._stubbed_seqs and not context._removed_seqs
+    # Let the withdrawn content age out of the ordinary recent-message window.
+    for i in range(30):
+        await context.add_message(
+            {"role": "assistant", "content": f"Later work {i} " + "x" * 200}
+        )
+    # A later request with a feasible requirement recovers normally.
+    view = await context.get_messages_for_request_retaining(retain_contents=[body])
+    assert any(m["content"] == body for m in view)
+    assert context._estimate_tokens(view) <= context.max_tokens
+
+
+@pytest.mark.asyncio
+async def test_post_compaction_failure_rolls_back_request_state_and_decisions():
+    context = SimpleContextManager(max_tokens=1000, compaction_notice_enabled=False)
+    required = "<system-reminders>Active retention requirement.</system-reminders>"
+    await context.add_message(human("First human request."))
+    await context.add_message(reminder(required))
+    await context.add_message(
+        {
+            "role": "assistant",
+            "content": "Unremovable loaded tool state: " + "x" * 5000,
+            "metadata": {"openai:tool_search_items": ["read_file"]},
+        }
+    )
+    await context.add_message(human("Latest human correction."))
+
+    with pytest.raises(ContextLengthError, match="Required content was not discarded"):
+        await context.get_messages_for_request_retaining(retain_contents=[required])
+
+    assert context._last_compaction_stats is None
+    assert context._last_token_meter_stats is None
+    assert not context._removed_seqs
+    assert not context._truncated_seqs
+    assert not context._stubbed_seqs
+    assert context._sticky_level == 0
+
+
+@pytest.mark.asyncio
+async def test_mount_advertises_additive_capability_and_keeps_old_getter():
+    class Coordinator:
+        def __init__(self):
+            self.capabilities = {}
+
+        async def mount(self, name, module):
+            self.context = module
+
+        def register_capability(self, name, capability):
+            self.capabilities[name] = capability
+
+    coordinator = Coordinator()
+    cleanup = await mount(coordinator)
+    assert callable(coordinator.capabilities["context.request_retention"])
+    assert await coordinator.context.get_messages_for_request() == []
+    assert (
+        await coordinator.capabilities["context.request_retention"](retain_contents=[])
+        == []
+    )
+    await cleanup()
+
+
+@pytest.mark.asyncio
+async def test_stale_actual_meter_cannot_prevent_feasible_retention_compaction():
+    context, body = await pressured_context()
+    context.token_meter = "actual"
+    context._last_measured_prompt_tokens = 100
+    view = await context.get_messages_for_request_retaining(retain_contents=[body])
+    assert any(m["content"] == body for m in view)
+    assert context._estimate_tokens(view) <= context.max_tokens
+
+
+@pytest.mark.asyncio
+async def test_changed_then_withdrawn_requirements_do_not_leave_a_stale_pin():
+    context = SimpleContextManager(max_tokens=1800, compaction_notice_enabled=False)
+    original = "<system-reminders>" + "original policy " * 180 + "</system-reminders>"
+    replacement = (
+        "<system-reminders>" + "replacement policy " * 180 + "</system-reminders>"
+    )
+    await context.add_message(human("First human request: preserve current requirements."))
+    await context.add_message(reminder(original))
+    for i in range(40):
+        await context.add_message(
+            {"role": "assistant", "content": f"Earlier work {i}: " + "x" * 1200}
+        )
+    await context.add_message(human("Latest human correction: use the replacement."))
+    await context.add_message(reminder(replacement))
+    for i in range(30):
+        await context.add_message(
+            {"role": "assistant", "content": f"Later work {i}: " + "z" * 500}
+        )
+
+    changed_view = await context.get_messages_for_request_retaining(
+        retain_contents=[replacement]
+    )
+    assert sum(m["content"] == original for m in changed_view) == 0
+    assert sum(m["content"] == replacement for m in changed_view) == 1
+
+    # A new, empty requirement set must permit the formerly selected envelope
+    # to age out on a real subsequent compaction.
+    for i in range(30):
+        await context.add_message(
+            {"role": "assistant", "content": f"Even later work {i}: " + "z" * 500}
+        )
+    withdrawn_view = await context.get_messages_for_request_retaining(
+        retain_contents=[]
+    )
+    assert not any(m["content"] == original for m in withdrawn_view)
+    assert not any(m["content"] == replacement for m in withdrawn_view)
+
+
+@pytest.mark.asyncio
+async def test_level_5_retains_complete_current_reminder_and_human_boundaries():
+    """A real Level 5 pass keeps the selected envelope and both human bodies."""
+    context = SimpleContextManager(
+        max_tokens=1800,
+        compact_threshold=0.92,
+        target_usage=0.50,
+        protected_recent=0.30,
+        compaction_notice_enabled=False,
+    )
+    first = human("First human request: preserve the complete acceptance criteria.")
+    active_reminder = reminder(
+        "<system-reminders>"
+        "Keep the current deployment region unchanged. "
+        "Do not use an unpublished credential. "
+        "Return the complete migration plan."
+        "</system-reminders>"
+    )
+    latest = human("Latest human correction: the migration must remain reversible.")
+    await context.add_message(first)
+    await context.add_message(active_reminder)
+    for i in range(14):
+        await context.add_message(
+            {"role": "assistant", "content": f"Machine output {i}: " + "x" * 800}
+        )
+    await context.add_message(latest)
+
+    canonical = copy.deepcopy(await context.get_messages())
+    first_envelope, reminder_envelope, *_, latest_envelope = map(
+        public_envelope, canonical
+    )
+    view = await context.get_messages_for_request_retaining(
+        retain_contents=[active_reminder["content"]]
+    )
+
+    assert context._last_compaction_stats is not None
+    assert context._last_compaction_stats["strategy_level"] == 5
+    assert first_envelope in view
+    assert reminder_envelope in view
+    assert latest_envelope in view
+    assert await context.get_messages() == canonical
+
+
+@pytest.mark.asyncio
+async def test_level_8_keeps_quoted_xml_human_prompt_and_reminder_complete():
+    """A quoted system-reminder is human content, not injected provenance."""
+    context = SimpleContextManager(
+        max_tokens=1800,
+        compact_threshold=0.92,
+        target_usage=0.50,
+        protected_recent=0.30,
+        compaction_notice_enabled=False,
+    )
+    first = human(
+        "<system-reminders>"
+        "A human quoted this XML while asking whether it was safe to publish."
+        "</system-reminders>"
+    )
+    active_reminder = reminder(
+        "<system-reminders>"
+        + "Active retention policy. " * 130
+        + "</system-reminders>"
+    )
+    latest = human("Latest human correction: do not remove the quoted evidence.")
+    await context.add_message(first)
+    await context.add_message(active_reminder)
+    for i in range(8):
+        await context.add_message(
+            {"role": "assistant", "content": f"Machine output {i}: " + "x" * 800}
+        )
+    await context.add_message(latest)
+
+    canonical = await context.get_messages()
+    first_envelope, reminder_envelope, *_, latest_envelope = map(
+        public_envelope, canonical
+    )
+    view = await context.get_messages_for_request_retaining(
+        retain_contents=[active_reminder["content"]]
+    )
+
+    assert context._last_compaction_stats is not None
+    assert context._last_compaction_stats["strategy_level"] == 8
+    assert first_envelope in view
+    assert reminder_envelope in view
+    assert latest_envelope in view
+    assert context._last_compaction_stats["after_tokens"] > context._last_compaction_stats[
+        "target_tokens"
+    ]
\ No newline at end of file

From 954bb545fd6d989e09ae848d791d1834dc2cc58a Mon Sep 17 00:00:00 2001
From: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
Date: Mon, 14 Sep 2026 13:30:17 -0700
Subject: [PATCH 2/4] docs: clarify request-retention scope

Generated with Amplifier

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
---
 README.md | 18 ++++++++++++++++--
 1 file changed, 16 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index 7fca880..185d362 100644
--- a/README.md
+++ b/README.md
@@ -99,11 +99,25 @@ Compaction triggers when token usage reaches the configured threshold (default:
 ### Protected Messages (Never Removed)
 
 - **System messages**: All system messages are always preserved
-- **First user message**: The original task/request is always protected (prevents losing context about what was originally asked)
-- **Last user message**: The most recent user input is always preserved
+- **First human prompt**: The original human task/request is protected, using message metadata rather than treating every user-role message as human input
+- **Last human prompt**: The most recent human input is protected by the same metadata-based classification
 - **Recent messages**: Last N% of messages (configurable via `protected_recent`)
 - **Tool pairs**: Tool_use and tool_result messages are treated as atomic units
 
+### Request-scoped retention
+
+The optional `context.request_retention` capability lets an orchestrator name
+the exact persisted reminder bodies required for its next request. The newest
+matching admitted `ephemeral=True, persisted=True` user-role envelope is kept
+complete through compaction, along with the first/latest human prompts.
+Quoting reminder XML in an ordinary human prompt does not change its identity.
+
+This protects delivery in the request view; it does not change message roles,
+pin every historical reminder, or rewrite canonical history. A missing required
+body or an irreducible required set that cannot fit raises `ContextLengthError`
+instead of silently dropping instructions. Failed assembly restores the prior
+compaction state.
+
 ### Compaction Phases
 
 1. **Phase 1 - Tool Result Truncation**: Older tool results are truncated to reduce token usage

From c36ff95f79137becffb0596be8ab9d143dcf3b5c Mon Sep 17 00:00:00 2001
From: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
Date: Mon, 14 Sep 2026 16:52:43 -0700
Subject: [PATCH 3/4] fix: preserve protected tool groups during compaction

Tested source bytes match the final-preflight3 guest per-file SHA-256 binding.

Generated with Amplifier

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
---
 README.md                                   |  4 ++
 amplifier_module_context_simple/__init__.py |  7 +++
 tests/test_progressive_compaction.py        | 50 +++++++++---------
 tests/test_protected_tool_results.py        | 56 ++++++++++++++++++++-
 4 files changed, 92 insertions(+), 25 deletions(-)

diff --git a/README.md b/README.md
index 185d362..a0f2d97 100644
--- a/README.md
+++ b/README.md
@@ -102,6 +102,7 @@ Compaction triggers when token usage reaches the configured threshold (default:
 - **First human prompt**: The original human task/request is protected, using message metadata rather than treating every user-role message as human input
 - **Last human prompt**: The most recent human input is protected by the same metadata-based classification
 - **Recent messages**: Last N% of messages (configurable via `protected_recent`)
+- **Recent tool results**: The last `protected_tool_results` results (default 5) are protected from both truncation and removal; a protected sibling also prevents removal of its owning call group
 - **Tool pairs**: Tool_use and tool_result messages are treated as atomic units
 
 ### Request-scoped retention
@@ -118,6 +119,9 @@ body or an irreducible required set that cannot fit raises `ContextLengthError`
 instead of silently dropping instructions. Failed assembly restores the prior
 compaction state.
 
+Protection takes precedence over the compaction target. A protected tool cohort
+can leave a view above that target; it is not a strict native-token ceiling.
+
 ### Compaction Phases
 
 1. **Phase 1 - Tool Result Truncation**: Older tool results are truncated to reduce token usage
diff --git a/amplifier_module_context_simple/__init__.py b/amplifier_module_context_simple/__init__.py
index 498bcbf..cfc8830 100644
--- a/amplifier_module_context_simple/__init__.py
+++ b/amplifier_module_context_simple/__init__.py
@@ -1790,6 +1790,13 @@ def _remove_messages_with_protection(
         for i in range(protected_boundary, len(messages)):
             protected_indices.add(i)
 
+        # The last N tool results are protected from removal as well as truncation.
+        # Their owning assistant and sibling results are vetoed atomically below.
+        tool_result_indices = [
+            i for i, msg in enumerate(messages) if msg.get("role") == "tool"
+        ]
+        protected_indices |= self._protected_tool_indices(tool_result_indices)
+
         # Removal candidates exclude ALL user messages (they can only be stubbed, not removed)
         removal_candidates = [
             i
diff --git a/tests/test_progressive_compaction.py b/tests/test_progressive_compaction.py
index ee433f5..3d3186f 100644
--- a/tests/test_progressive_compaction.py
+++ b/tests/test_progressive_compaction.py
@@ -15,18 +15,15 @@
 @pytest.mark.asyncio
 async def test_tool_result_truncation_phase1():
     """Phase 1 truncates old tool results while preserving structure."""
-    # Configure for easy testing: low tokens, aggressive truncation
-    # Use lower max_tokens to ensure compaction triggers with our message sizes
+    # Four results make the first 25% truncation wave nonempty. Its large
+    # result is unprotected; the three recent results must remain intact.
     context = SimpleContextManager(
-        max_tokens=300,  # Lower to ensure compaction triggers
-        compact_threshold=0.5,
-        target_usage=0.3,
+        max_tokens=2000,
+        compact_threshold=0.9,
+        target_usage=0.5,
         truncate_chars=50,
-        protected_recent=0.3,  # Protect recent messages but allow truncation of old tool
-        protected_tool_results=2,  # Only protect last 2 tool results
-        # Without this, the default 800-token notice reserve makes
-        # effective_budget = 300 - 800 = -700, which silently disables
-        # compaction entirely (see test_budget_guard.py).
+        protected_recent=0,
+        protected_tool_results=3,
         compaction_notice_enabled=False,
     )
 
@@ -46,19 +43,22 @@ async def test_tool_result_truncation_phase1():
         }
     )
     # Add a large tool result - this will push us over the threshold
-    large_content = "x" * 500  # 500 chars = ~125 tokens
+    large_content = "x" * 10000
     await context.add_message(
         {"role": "tool", "tool_call_id": "toolu_early", "content": large_content}
     )
 
-    # Add more messages to push tool pair into truncate zone (first 50%)
-    for i in range(10):
+    for i in range(3):
+        call_id = f"recent-call-{i}"
         await context.add_message(
-            {"role": "user", "content": f"message {i} with extra padding"}
+            {"role": "assistant", "content": "", "tool_calls": [
+                {"id": call_id, "type": "function", "function": {"name": "read_file"}}
+            ]}
         )
         await context.add_message(
-            {"role": "assistant", "content": f"response {i} with extra padding"}
+            {"role": "tool", "tool_call_id": call_id, "content": "small result"}
         )
+    await context.add_message({"role": "user", "content": "continue"})
 
     # Trigger compaction
     messages = await context.get_messages_for_request()
@@ -70,14 +70,18 @@ async def test_tool_result_truncation_phase1():
             tool_result = msg
             break
 
-    # Verify truncation occurred
-    if tool_result:
-        content = tool_result.get("content", "")
-        assert "[truncated:" in content, "Tool result should be truncated"
-        assert tool_result.get("_truncated") is True, "Should have _truncated marker"
-        assert len(content) < 200, (
-            f"Truncated content should be small, got {len(content)}"
-        )
+    # Removal is not a substitute for the truncation this test promises.
+    assert tool_result is not None
+    content = tool_result.get("content", "")
+    assert "[truncated:" in content, "Tool result should be truncated"
+    assert tool_result.get("_truncated") is True
+    assert len(content) < 200
+    assert context._last_compaction_stats["strategy_level"] == 1
+    assert context._last_compaction_stats["messages_removed"] == 0
+    assert [
+        m["content"] for m in messages
+        if m.get("tool_call_id", "").startswith("recent-call-")
+    ] == ["small result"] * 3
 
 
 @pytest.mark.asyncio
diff --git a/tests/test_protected_tool_results.py b/tests/test_protected_tool_results.py
index 2312b4b..0bb9eda 100644
--- a/tests/test_protected_tool_results.py
+++ b/tests/test_protected_tool_results.py
@@ -164,8 +164,8 @@ async def test_protecting_every_tool_result_still_works():
     result = await _run_workload(_make_context(protected_tool_results=N_TOOL_PAIRS))
 
     assert result["truncated_tool_ids"] == [], result
-    assert result["level"] == 3, result
-    assert result["messages_removed"] > 0, result
+    assert result["level"] >= 3, result
+    assert result["messages_removed"] == 0, result
 
 
 # --------------------------------------------------------------------------
@@ -201,3 +201,55 @@ def test_protected_tool_indices_uses_real_positions_not_ordinals():
     """The returned indices are positions in the message list, not 0..N-1."""
     ctx = SimpleContextManager(protected_tool_results=2)
     assert ctx._protected_tool_indices([3, 11, 40, 57]) == {40, 57}
+
+
+@pytest.mark.parametrize("protected", [0, 1, 5])
+def test_one_protected_sibling_vetoes_whole_batch_removal(protected):
+    """Removal honors the same last-N floor as truncation, including N=0."""
+    from copy import deepcopy
+
+    ctx = SimpleContextManager(protected_tool_results=protected)
+    messages = [
+        {"role": "user", "content": "first human"},
+        {"role": "assistant", "content": "", "thinking": {"opaque": "unchanged"},
+         "tool_calls": [
+             {"id": "old", "type": "function", "function": {"name": "read_file"}},
+             {"id": "recent", "type": "function", "function": {"name": "read_file"}},
+         ]},
+        {"role": "tool", "tool_call_id": "old", "content": "older sibling"},
+        {"role": "tool", "tool_call_id": "recent", "content": "protected sibling"},
+        {"role": "user", "content": "last human"},
+    ]
+    before = deepcopy(messages)
+    view, removed, stubbed, _ = ctx._remove_messages_with_protection(
+        messages, target_tokens=1, protected_recent=0, system_tokens=0,
+    )
+    assert messages == before
+    assert stubbed == 0
+    assert removed == (0 if protected else 3)
+    assert view == (before if protected else [before[0], before[-1]])
+
+
+@pytest.mark.asyncio
+async def test_default_tool_floor_survives_progressive_removal():
+    from copy import deepcopy
+
+    ctx = SimpleContextManager(max_tokens=120_000)
+    for index in range(4):
+        await ctx.add_message({"role": "user", "content": f"human-{index}:" + "x" * 100_000})
+        await ctx.add_message({
+            "role": "assistant", "content": "", "tool_calls": [
+                {"id": f"call-{index}", "type": "function", "function": {"name": "read_file"}}
+            ],
+        })
+        await ctx.add_message({
+            "role": "tool", "tool_call_id": f"call-{index}", "content": f"result-{index}",
+        })
+    await ctx.add_message({"role": "user", "content": "latest:" + "y" * 100_000})
+    canonical = deepcopy(ctx.messages)
+    view = await ctx.get_messages_for_request()
+    assert ctx._last_compaction_stats["strategy_level"] >= 4
+    assert [m["content"] for m in view if m.get("role") == "tool"] == [
+        f"result-{index}" for index in range(4)
+    ]
+    assert ctx.messages == canonical

From 6a087e3a6747f34a9ff9a6d20a1149ca54e5a758 Mon Sep 17 00:00:00 2001
From: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
Date: Mon, 14 Sep 2026 18:20:16 -0700
Subject: [PATCH 4/4] ci: test context retention against released Core 1.6.1

Generated with Amplifier

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
---
 .github/workflows/ci.yml |  8 ++++++--
 README.md                | 13 ++++++++++++-
 pyproject.toml           |  5 +----
 uv.lock                  | 14 +++++++++++---
 4 files changed, 30 insertions(+), 10 deletions(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index e41a6ef..bca3e31 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -62,7 +62,11 @@ jobs:
           python-version: ${{ matrix.python-version }}
 
       - name: Install dependencies
-        run: uv sync --all-extras --dev
+        run: uv sync --locked --all-extras --dev
+
+      - name: Verify released Core API
+        run: |
+          uv run --locked python -c "from importlib.metadata import version; from amplifier_core.llm_errors import ContextLengthError; assert version('amplifier-core') == '1.6.1'"
 
       # No API key placeholder here, unlike provider-openai's CI. Measured:
       # with ANTHROPIC_API_KEY/OPENAI_API_KEY unset, the suite is
@@ -72,4 +76,4 @@ jobs:
       # No `-m "not live"` deselection either: this repo registers no `live`
       # marker and has no network-dependent tests.
       - name: Run test suite
-        run: uv run pytest -q
+        run: uv run --locked pytest -q
diff --git a/README.md b/README.md
index a0f2d97..4282ddd 100644
--- a/README.md
+++ b/README.md
@@ -297,7 +297,18 @@ a corresponding quality regression.
 
 ## Dependencies
 
-- `amplifier-core>=1.0.0`
+- Host-provided `amplifier-core>=1.6.1`, including
+  `amplifier_core.llm_errors.ContextLengthError` for fail-loud request retention.
+  Core remains provided by the host, not installed as a runtime module dependency.
+
+Development and CI pin the released `amplifier-core==1.6.1` package in `uv.lock`.
+A Git `main` source mapping can retain an older commit in the lockfile; the
+release pin ensures tests exercise the Core API required by this module.
+
+```bash
+uv sync --locked --all-extras --dev
+uv run --locked pytest -q
+```
 
 ## Contributing
 
diff --git a/pyproject.toml b/pyproject.toml
index 50d7020..8a7dd4f 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -30,11 +30,8 @@ packages = [
 [tool.hatch.metadata]
 allow-direct-references = true
 
-[tool.uv.sources]
-amplifier-core = { git = "https://github.com/microsoft/amplifier-core", branch = "main" }
-
 [dependency-groups]
-dev = ["amplifier-core", "pytest>=9.0.3", "pytest-asyncio>=0.23.0"]
+dev = ["amplifier-core==1.6.1", "pytest>=9.0.3", "pytest-asyncio>=0.23.0"]
 
 [tool.pytest.ini_options]
 testpaths = ["tests"]
diff --git a/uv.lock b/uv.lock
index 0f2e81c..3223a56 100644
--- a/uv.lock
+++ b/uv.lock
@@ -4,8 +4,8 @@ requires-python = ">=3.11"
 
 [[package]]
 name = "amplifier-core"
-version = "1.0.0"
-source = { git = "https://github.com/microsoft/amplifier-core?branch=main#976fb87335ffe398cf0c1bd1bcd3c2f2c154fc3c" }
+version = "1.6.1"
+source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "click" },
     { name = "pydantic" },
@@ -13,6 +13,14 @@ dependencies = [
     { name = "tomli" },
     { name = "typing-extensions" },
 ]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/03/cd/8b0b520bf0de741ea73e069aaf64aca28c9f4ce91a7b8b9239193a6c4c1b/amplifier_core-1.6.1-cp311-abi3-macosx_10_12_x86_64.whl", hash = "sha256:c0f711d8408de78e53e5deddcb38b7240c5c1c497ca51eeaaeff23559b3d3c48", size = 8281633, upload-time = "2026-08-10T02:38:11.98Z" },
+    { url = "https://files.pythonhosted.org/packages/14/83/f4fb297d87d35b9d74058da02bb153e12f7891ab62b3aaf7e0857f877798/amplifier_core-1.6.1-cp311-abi3-macosx_11_0_arm64.whl", hash = "sha256:b08f37e2c0b1611349a0e25d5bf9bfdfae3afcee35488f8e26bba1cdd400503b", size = 7366930, upload-time = "2026-08-10T02:38:14.105Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/ba/5eb9cecf92d8053c5e6d46ad9668c3ed3558d5423845c1dced1f266b2a38/amplifier_core-1.6.1-cp311-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6ebf7e3993c76ea506e70ac7844b286c3ba2e9127b3bcb350fa4fcd2dcdbd38d", size = 7659512, upload-time = "2026-08-10T02:38:16.314Z" },
+    { url = "https://files.pythonhosted.org/packages/22/31/121f054e3d079dc33d83f3d8ba9af50fd9f7694c3e2ba3d7d23d7c157d48/amplifier_core-1.6.1-cp311-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3c957cd0671d2a003f2c8f7d6a41bd6e808f97d183c57b97e7700bf4c912621d", size = 8678425, upload-time = "2026-08-10T02:38:18.243Z" },
+    { url = "https://files.pythonhosted.org/packages/35/25/bfc217f4a9ed2d033995fc59847f1fee2e1b17130632fcb0e0981a1a311b/amplifier_core-1.6.1-cp311-abi3-win_amd64.whl", hash = "sha256:50c80bcfa1f6efe769b19e7af18c925024c7553d4db08880727241709dd44eae", size = 8976601, upload-time = "2026-08-10T02:38:20.505Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/14/5f330452c92c6c5d35c51ad5311301949ce5db4d1a1a901456f3ee43eaac/amplifier_core-1.6.1-cp311-abi3-win_arm64.whl", hash = "sha256:cd8b617f132cf5d1ca3e5187d5f831d1f2a508bb40d07b2ab1085961bcb9e1a9", size = 7744837, upload-time = "2026-08-10T02:38:22.562Z" },
+]
 
 [[package]]
 name = "amplifier-module-context-simple"
@@ -30,7 +38,7 @@ dev = [
 
 [package.metadata.requires-dev]
 dev = [
-    { name = "amplifier-core", git = "https://github.com/microsoft/amplifier-core?branch=main" },
+    { name = "amplifier-core", specifier = "==1.6.1" },
     { name = "pytest", specifier = ">=9.0.3" },
     { name = "pytest-asyncio", specifier = ">=0.23.0" },
 ]