fix: strip SAAR markers from team_rules -- prevents inception loop and triple duplication

Author: DevanshuNEU

CLAUDE.md

@@ -1,7 +1,7 @@
 <!-- SAAR:AUTO-START -->
 # CLAUDE.md -- saar
 
-252 functions, 38 classes.
+253 functions, 38 classes.
 Async adoption: 21%.
 Type hint coverage: 92%.
 
@@ -35,11 +35,11 @@ import os
 These files have the most dependents -- understand them before editing:
 
 - `saar/models.py` (14 dependents)
+- `saar/formatters/agents_md.py` (4 dependents)
 - `saar/formatters/_tribal.py` (4 dependents)
-- `saar/interview.py` (3 dependents)
 - `saar/cli.py` (3 dependents)
-- `saar/formatters/agents_md.py` (3 dependents)
 - `saar/formatters/claude_md.py` (3 dependents)
+- `saar/interview.py` (3 dependents)
 - `saar/dependency_analyzer.py` (2 dependents)
 - `saar/style_analyzer.py` (2 dependents)
 

saar/formatters/agents_md.py

@@ -12,6 +12,46 @@
 from saar.models import CodebaseDNA
 from saar.formatters._tribal import render_tribal_knowledge
 
+# Markers we strip from team_rules before embedding -- prevents nested markers
+# and the inception loop where saar reads its own output as input
+_SAAR_MARKER_START = "<!-- SAAR:AUTO-START -->"
+_SAAR_MARKER_END = "<!-- SAAR:AUTO-END -->"
+
+# Max chars to include from hand-written team rules.
+# The auto-detected DNA above already covers conventions -- team rules should
+# add what's MISSING (data flow, key services, commit rules), not repeat them.
+_TEAM_RULES_MAX_CHARS = 3000
+
+
+def _clean_team_rules(raw: str) -> str:
+    """Strip SAAR markers and truncate team rules before embedding.
+
+    Why: team_rules comes from reading CLAUDE.md or .cursorrules verbatim.
+    If CLAUDE.md was previously generated by saar, it contains SAAR:AUTO-START
+    markers, causing nested markers and triple duplication on re-runs.
+
+    We strip the auto-generated block (between markers) and keep only the
+    human-written sections that live outside the markers.
+    """
+    # strip auto-generated blocks between markers -- those are already
+    # covered by the DNA sections above, no need to repeat them
+    result = raw
+    while _SAAR_MARKER_START in result and _SAAR_MARKER_END in result:
+        start = result.find(_SAAR_MARKER_START)
+        end = result.find(_SAAR_MARKER_END, start)
+        if start == -1 or end == -1:
+            break
+        result = result[:start] + result[end + len(_SAAR_MARKER_END):]
+
+    result = result.strip()
+
+    # truncate if still very long -- human rules should be concise
+    if len(result) > _TEAM_RULES_MAX_CHARS:
+        result = result[:_TEAM_RULES_MAX_CHARS].rstrip()
+        result += "\n\n*(truncated -- see full file for remaining rules)*"
+
+    return result
+
 
 def render_agents_md(dna: CodebaseDNA) -> str:
     """Render DNA as an AGENTS.md file."""
@@ -164,11 +204,13 @@ def render_agents_md(dna: CodebaseDNA) -> str:
     if tribal:
         lines.append(tribal)
 
-    # -- team rules verbatim --
+    # -- team rules verbatim (human-written sections only) --
     if dna.team_rules:
-        lines.append("\n## Project-Specific Rules\n")
-        if dna.team_rules_source:
-            lines.append(f"*From `{dna.team_rules_source}`*\n")
-        lines.append(dna.team_rules)
+        cleaned = _clean_team_rules(dna.team_rules)
+        if cleaned:
+            lines.append("\n## Project-Specific Rules\n")
+            if dna.team_rules_source:
+                lines.append(f"*From `{dna.team_rules_source}`*\n")
+            lines.append(cleaned)
 
     return "\n".join(lines) + "\n"

saar/formatters/claude_md.py

@@ -6,6 +6,7 @@
 """
 from saar.models import CodebaseDNA
 from saar.formatters._tribal import render_tribal_knowledge
+from saar.formatters.agents_md import _clean_team_rules
 
 
 def render_claude_md(dna: CodebaseDNA) -> str:
@@ -139,11 +140,13 @@ def render_claude_md(dna: CodebaseDNA) -> str:
     if tribal:
         lines.append(tribal)
 
-    # -- team rules (verbatim if they exist) --
+    # -- team rules (human-written sections only) --
     if dna.team_rules:
-        lines.append("\n## Team Rules\n")
-        if dna.team_rules_source:
-            lines.append(f"*Imported from `{dna.team_rules_source}`*\n")
-        lines.append(dna.team_rules)
+        cleaned = _clean_team_rules(dna.team_rules)
+        if cleaned:
+            lines.append("\n## Team Rules\n")
+            if dna.team_rules_source:
+                lines.append(f"*Imported from `{dna.team_rules_source}`*\n")
+            lines.append(cleaned)
 
     return "\n".join(lines) + "\n"