feat(compression): make protect_first_n configurable

The number of head messages preserved verbatim across context compactions
was previously hardcoded to 3 in AIAgent.__init__. Expose it as
compression.protect_first_n in config, matching the existing
protect_last_n pattern.

Motivation: users who rely on rolling compaction for long-running
sessions had the opening user/assistant exchange pinned as head
forever, which doesn't always match how they want the session framed
after many compactions. Lowering to 1 preserves only the system
prompt and lets the first exchange age out naturally through the
summary.

- hermes_cli/config.py: add default (3) + show_config() line
- run_agent.py: read from config with max(1, ...) floor so the
  system prompt always survives head-protection
- cli-config.yaml.example: document with rationale
- tests/agent/test_context_compressor.py: cover default, override,
  and end-to-end behavior when protect_first_n=1
- scripts/release.py: add simpolism@gmail.com to AUTHOR_MAP

cli-config.yaml.example

@@ -323,6 +323,16 @@ compression:
   # compression of older turns.
   protect_last_n: 20
 
+  # Number of earliest messages to always preserve verbatim as head (default: 3 =
+  # system prompt + first user/assistant exchange). Head messages are NEVER
+  # summarized — they survive every compression indefinitely. This gives stable
+  # early context for short/medium sessions, but in long-running sessions that
+  # rely on rolling compaction the opening user turn gets pinned forever, which
+  # may not match how you want the session framed. Lower to 1 to preserve only
+  # the system prompt and let the first exchange age out naturally through the
+  # summary. Minimum: 1 (system prompt must survive).
+  protect_first_n: 3
+
   # To pin a specific model/provider for compression summaries, use the
   # auxiliary section below (auxiliary.compression.provider / model).
 

hermes_cli/config.py

@@ -473,6 +473,11 @@ DEFAULT_CONFIG = {
         "threshold": 0.50,            # compress when context usage exceeds this ratio
         "target_ratio": 0.20,         # fraction of threshold to preserve as recent tail
         "protect_last_n": 20,         # minimum recent messages to keep uncompressed
+        "protect_first_n": 3,         # head messages always preserved verbatim
+                                      # (system prompt + first exchange). Lower this
+                                      # to 1 if you rely on rolling compaction for
+                                      # long-running sessions and don't want the
+                                      # opening user/assistant turn pinned forever.
 
     },
 
@@ -3599,6 +3604,7 @@ def show_config():
         print(f"  Threshold:    {compression.get('threshold', 0.50) * 100:.0f}%")
         print(f"  Target ratio: {compression.get('target_ratio', 0.20) * 100:.0f}% of threshold preserved")
         print(f"  Protect last: {compression.get('protect_last_n', 20)} messages")
+        print(f"  Protect first: {compression.get('protect_first_n', 3)} messages")
         _aux_comp = config.get('auxiliary', {}).get('compression', {})
         _sm = _aux_comp.get('model', '') or '(auto)'
         print(f"  Model:        {_sm}")

run_agent.py

@@ -1538,6 +1538,12 @@ class AIAgent:
         compression_enabled = str(_compression_cfg.get("enabled", True)).lower() in ("true", "1", "yes")
         compression_target_ratio = float(_compression_cfg.get("target_ratio", 0.20))
         compression_protect_last = int(_compression_cfg.get("protect_last_n", 20))
+        # Floor protect_first_n at 1 — the system prompt must always be
+        # preserved as head, otherwise compression would strip it on the
+        # first compaction pass and the model would lose its identity.
+        compression_protect_first = max(
+            1, int(_compression_cfg.get("protect_first_n", 3))
+        )
 
         # Read optional explicit context_length override for the auxiliary
         # compression model. Custom endpoints often cannot report this via
@@ -1684,7 +1690,7 @@ class AIAgent:
             self.context_compressor = ContextCompressor(
                 model=self.model,
                 threshold_percent=compression_threshold,
-                protect_first_n=3,
+                protect_first_n=compression_protect_first,
                 protect_last_n=compression_protect_last,
                 summary_target_ratio=compression_target_ratio,
                 summary_model_override=None,

scripts/release.py

@@ -183,6 +183,7 @@ AUTHOR_MAP = {
     "m@statecraft.systems": "mbierling",
     "balyan.sid@gmail.com": "alt-glitch",
     "oluwadareab12@gmail.com": "bennytimz",
+    "simpolism@gmail.com": "simpolism",
     "simon@simonmarcus.org": "simon-marcus",
     "xowiekk@gmail.com": "Xowiek",
     "1243352777@qq.com": "zons-zhaozhy",

tests/agent/test_context_compressor.py

@@ -627,6 +627,49 @@ class TestSummaryTargetRatio:
             c = ContextCompressor(model="test", quiet_mode=True)
         assert c.protect_last_n == 20
 
+    def test_default_protect_first_n_is_3(self):
+        """Default protect_first_n should be 3 (system prompt + first exchange)."""
+        with patch("agent.context_compressor.get_model_context_length", return_value=100_000):
+            c = ContextCompressor(model="test", quiet_mode=True)
+        assert c.protect_first_n == 3
+
+    def test_protect_first_n_override(self):
+        """protect_first_n=1 should be honoured — for users who rely on rolling
+        compaction and don't want the opening user/assistant turn pinned as head
+        indefinitely.  Only the system prompt survives head-protection."""
+        with patch("agent.context_compressor.get_model_context_length", return_value=100_000):
+            c = ContextCompressor(model="test", quiet_mode=True, protect_first_n=1)
+        assert c.protect_first_n == 1
+
+    def test_protect_first_n_1_preserves_only_system_prompt(self):
+        """End-to-end: when protect_first_n=1, compression should treat only
+        the first message (system prompt) as head.  Messages 1..n-protect_last-1
+        become summarization candidates, unlike the default where messages 0-2
+        would all be pinned as head."""
+        with patch("agent.context_compressor.get_model_context_length", return_value=100_000):
+            c = ContextCompressor(
+                model="test",
+                quiet_mode=True,
+                protect_first_n=1,
+                protect_last_n=2,
+            )
+        msgs = (
+            [{"role": "system", "content": "System prompt"}]
+            + [{"role": "user" if i % 2 == 0 else "assistant", "content": f"msg {i}"}
+               for i in range(8)]
+        )
+        result = c.compress(msgs)
+        # System prompt (msg[0]) survives as head
+        assert result[0]["role"] == "system"
+        assert result[0]["content"].startswith("System prompt")
+        # The first user/assistant exchange (msg 0, msg 1) should NOT be pinned
+        # as head verbatim — those would have been summarized or absorbed.
+        # Under default protect_first_n=3, result[1] and result[2] would be
+        # the literal "msg 0" / "msg 1"; with protect_first_n=1 they aren't.
+        assert result[1].get("content") != "msg 0"
+        # Last 2 messages are tail-protected under protect_last_n=2
+        assert result[-1]["content"] == msgs[-1]["content"]
+
 
 class TestTokenBudgetTailProtection:
     """Tests for token-budget-based tail protection (PR #6240).