From df1bbdf92471977901af19651dca5c578bc745d9 Mon Sep 17 00:00:00 2001
From: terp <terp@local>
Date: Wed, 27 May 2026 12:32:31 -0400
Subject: [PATCH] docs: fix two code example accuracy issues

Re-apply the valid parts of #4 on top of current main:

- part6: the fallback branch in the compression example now assigns
  compressed_context = messages_to_compress instead of returning
  undefined original_context, matching the surrounding assignment style.
- part11: the gateway health-check script now uses df -Ph so awk NR==2
  is stable even when long device names would make df -h wrap lines.

Intentionally does not apply #4's part7 change. Current Hermes source
(/home/terp/.hermes/hermes-agent/tools/session_search_tool.py) explicitly
states that session_search has no LLM calls and no summary LLM path, so
the existing no-API-call wording remains accurate.
---
 part11-gateway-recovery.md   | 2 +-
 part6-context-compression.md | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/part11-gateway-recovery.md b/part11-gateway-recovery.md
index 83af1d9..8f9f1c1 100644
--- a/part11-gateway-recovery.md
+++ b/part11-gateway-recovery.md
@@ -187,7 +187,7 @@ if ! curl -s http://localhost:8642/health > /dev/null 2>&1; then
 fi
 
 # Disk space OK?
-USAGE=$(df -h ~/.hermes | awk 'NR==2 {print $5}' | tr -d '%')
+USAGE=$(df -Ph ~/.hermes | awk 'NR==2 {print $5}' | tr -d '%')
 if [ "$USAGE" -gt 90 ]; then
     echo "WARNING: Disk usage at ${USAGE}%"
     exit 1
diff --git a/part6-context-compression.md b/part6-context-compression.md
index 7639968..48ba380 100644
--- a/part6-context-compression.md
+++ b/part6-context-compression.md
@@ -47,7 +47,7 @@ try:
     compressed_context = summary
 except Exception as e:
     logger.warning(f"Context compression failed: {e}, preserving original context")
-    return original_context  # Don't compress, don't lose data
+    compressed_context = messages_to_compress  # Don't compress, don't lose data
 ```
 
 **The rule:** If compression can't succeed, keep the uncompressed context. A slower response is better than a wrong one.