Configs changes

.env.example

@@ -125,6 +125,7 @@ DEFAULT_MODEL=anthropic/claude-opus-4-6
 # EXTRACTION_MODEL=anthropic/claude-sonnet-4-5-20250929
 
 # DEFAULT_MAX_ITERS=10
+# DEFAULT_MAX_LLM_CALLS=30
 # Provider reasoning budget: minimal | low | medium | high
 # DEFAULT_REASONING_EFFORT=medium
 # DEFAULT_TEMPERATURE=0.2
@@ -151,6 +152,9 @@ DEFAULT_MODEL=anthropic/claude-opus-4-6
 # Enable provider-side prompt caching (reduces latency and costs)
 # ENABLE_PROMPT_CACHING=true
 
+# Minimum confidence threshold for reported issues (0.0-1.0)
+# MIN_CONFIDENCE=0.81
+
 # =============================================================================
 # Memory (Hippocampus)
 # =============================================================================
@@ -250,6 +254,7 @@ DEFAULT_MODEL=anthropic/claude-opus-4-6
 # Available settings per signature:
 #   - ENABLED (true/false)
 #   - MAX_ITERS (integer)
+#   - MAX_LLM_CALLS (integer) - RLM fallback max calls (unset -> DEFAULT_MAX_LLM_CALLS)
 #   - MODEL (LiteLLM model string)
 #   - REASONING_EFFORT (minimal|low|medium|high) - Provider reasoning budget
 #   - TEMPERATURE (float) - Must be 1 while reasoning is enabled

CHANGELOG.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## [1.0.3] - 2026-08-18
+
+### Added
+- Configurable `DEFAULT_MAX_LLM_CALLS` setting (default 30) controlling maximum LLM calls for RLM context-overflow fallback
+- Configurable `MIN_CONFIDENCE` threshold (default 0.81) for filtering low-confidence issues
+- Per-signature `max_llm_calls` override (`SCOPE_MAX_LLM_CALLS`, `CODE_REVIEW_MAX_LLM_CALLS`, `SUPPLY_CHAIN_MAX_LLM_CALLS`, `SUMMARY_MAX_LLM_CALLS`, `DOC_MAX_LLM_CALLS`, `AUDIT_MAX_LLM_CALLS`)
+- `ContextSafe` now accepts `max_iters` and `max_llm_calls` parameters for per-module override
+
+### Changed
+- `default-max-iters` action input default raised from 3 to 10
+- Moved hardcoded `MIN_CONFIDENCE` constant from `helpers.py` to configurable `Settings.min_confidence` field (default 0.81)
+
+### Fixed
+- RLM fallback previously hardcoded `max_iters=10, max_llm_calls=20`; now respects per-signature and global configuration
+
 ## [1.0.2] - 2026-08-18
 
 ### Fixed

action.yml

@@ -66,7 +66,7 @@ inputs:
   default-max-iters:
     description: 'Default maximum iterations for all signatures'
     required: false
-    default: '3'
+    default: '10'
 
   default-reasoning-effort:
     description: 'Default reasoning effort for all signatures (minimal|low|medium|high)'
@@ -93,7 +93,17 @@ inputs:
     description: 'Enable provider-side prompt caching (reduces latency and costs)'
     required: false
     default: 'true'
-
+
+  default-max-llm-calls:
+    description: 'Maximum LLM calls for RLM fallback (context overflow recovery)'
+    required: false
+    default: '30'
+
+  min-confidence:
+    description: 'Minimum confidence threshold (0.0-1.0) for reported issues'
+    required: false
+    default: '0.81'
+
   # ==========================================
   # SIGNATURE: scope
   # ==========================================
@@ -109,7 +119,11 @@ inputs:
   scope-max-iters:
     description: 'Max iterations for scope identification'
     required: false
-
+
+  scope-max-llm-calls:
+    description: 'Max LLM calls for scope RLM fallback'
+    required: false
+
   scope-reasoning-effort:
     description: 'Reasoning effort for scope (minimal|low|medium|high)'
     required: false
@@ -133,7 +147,11 @@ inputs:
   code-review-max-iters:
     description: 'Max iterations for code review'
     required: false
-
+
+  code-review-max-llm-calls:
+    description: 'Max LLM calls for code review RLM fallback'
+    required: false
+
   code-review-reasoning-effort:
     description: 'Reasoning effort for code review (minimal|low|medium|high)'
     required: false
@@ -177,7 +195,11 @@ inputs:
   supply-chain-max-iters:
     description: 'Max iterations for supply chain security'
     required: false
-
+
+  supply-chain-max-llm-calls:
+    description: 'Max LLM calls for supply chain RLM fallback'
+    required: false
+
   supply-chain-reasoning-effort:
     description: 'Reasoning effort for supply chain (minimal|low|medium|high)'
     required: false
@@ -203,6 +225,10 @@ inputs:
     description: 'Model for summary (empty = use default)'
     required: false
 
+  summary-max-llm-calls:
+    description: 'Max LLM calls for summary RLM fallback'
+    required: false
+
   summary-reasoning-effort:
     description: 'Reasoning effort for summary (minimal|low|medium|high)'
     required: false
@@ -361,18 +387,22 @@ runs:
         LLM_RETRIES: ${{ inputs.llm-retries }}
         LLM_TIMEOUT: ${{ inputs.llm-timeout }}
         ENABLE_PROMPT_CACHING: ${{ inputs.enable-prompt-caching }}
-
+        DEFAULT_MAX_LLM_CALLS: ${{ inputs.default-max-llm-calls }}
+        MIN_CONFIDENCE: ${{ inputs.min-confidence }}
+
         # Scope identification signature
         SCOPE_ENABLED: ${{ inputs.scope-enabled }}
         SCOPE_MODEL: ${{ inputs.scope-model }}
         SCOPE_MAX_ITERS: ${{ inputs.scope-max-iters }}
+        SCOPE_MAX_LLM_CALLS: ${{ inputs.scope-max-llm-calls }}
         SCOPE_REASONING_EFFORT: ${{ inputs.scope-reasoning-effort }}
         SCOPE_TEMPERATURE: ${{ inputs.scope-temperature }}
 
         # Code review signature
         CODE_REVIEW_ENABLED: ${{ inputs.code-review-enabled }}
         CODE_REVIEW_MODEL: ${{ inputs.code-review-model }}
         CODE_REVIEW_MAX_ITERS: ${{ inputs.code-review-max-iters }}
+        CODE_REVIEW_MAX_LLM_CALLS: ${{ inputs.code-review-max-llm-calls }}
         CODE_REVIEW_REASONING_EFFORT: ${{ inputs.code-review-reasoning-effort }}
         CODE_REVIEW_TEMPERATURE: ${{ inputs.code-review-temperature }}
 
@@ -386,13 +416,15 @@ runs:
         SUPPLY_CHAIN_ENABLED: ${{ inputs.supply-chain-enabled }}
         SUPPLY_CHAIN_MODEL: ${{ inputs.supply-chain-model }}
         SUPPLY_CHAIN_MAX_ITERS: ${{ inputs.supply-chain-max-iters }}
+        SUPPLY_CHAIN_MAX_LLM_CALLS: ${{ inputs.supply-chain-max-llm-calls }}
         SUPPLY_CHAIN_REASONING_EFFORT: ${{ inputs.supply-chain-reasoning-effort }}
         SUPPLY_CHAIN_TEMPERATURE: ${{ inputs.supply-chain-temperature }}
         SUPPLY_CHAIN_SCAN_UNCHANGED: ${{ inputs.supply-chain-scan-unchanged }}
 
         # Summary signature
         SUMMARY_ENABLED: ${{ inputs.summary-enabled }}
         SUMMARY_MODEL: ${{ inputs.summary-model }}
+        SUMMARY_MAX_LLM_CALLS: ${{ inputs.summary-max-llm-calls }}
         SUMMARY_REASONING_EFFORT: ${{ inputs.summary-reasoning-effort }}
         SUMMARY_TEMPERATURE: ${{ inputs.summary-temperature }}
 
@@ -441,18 +473,22 @@ runs:
         [ -n "$LLM_RETRIES" ] && DOCKER_ARGS="$DOCKER_ARGS -e LLM_RETRIES"
         [ -n "$LLM_TIMEOUT" ] && DOCKER_ARGS="$DOCKER_ARGS -e LLM_TIMEOUT"
         [ -n "$ENABLE_PROMPT_CACHING" ] && DOCKER_ARGS="$DOCKER_ARGS -e ENABLE_PROMPT_CACHING"
-
+        [ -n "$DEFAULT_MAX_LLM_CALLS" ] && DOCKER_ARGS="$DOCKER_ARGS -e DEFAULT_MAX_LLM_CALLS"
+        [ -n "$MIN_CONFIDENCE" ] && DOCKER_ARGS="$DOCKER_ARGS -e MIN_CONFIDENCE"
+
         # Scope identification
         [ -n "$SCOPE_ENABLED" ] && DOCKER_ARGS="$DOCKER_ARGS -e SCOPE_ENABLED"
         [ -n "$SCOPE_MODEL" ] && DOCKER_ARGS="$DOCKER_ARGS -e SCOPE_MODEL"
         [ -n "$SCOPE_MAX_ITERS" ] && DOCKER_ARGS="$DOCKER_ARGS -e SCOPE_MAX_ITERS"
+        [ -n "$SCOPE_MAX_LLM_CALLS" ] && DOCKER_ARGS="$DOCKER_ARGS -e SCOPE_MAX_LLM_CALLS"
         [ -n "$SCOPE_REASONING_EFFORT" ] && DOCKER_ARGS="$DOCKER_ARGS -e SCOPE_REASONING_EFFORT"
         [ -n "$SCOPE_TEMPERATURE" ] && DOCKER_ARGS="$DOCKER_ARGS -e SCOPE_TEMPERATURE"
 
         # Code review
         [ -n "$CODE_REVIEW_ENABLED" ] && DOCKER_ARGS="$DOCKER_ARGS -e CODE_REVIEW_ENABLED"
         [ -n "$CODE_REVIEW_MODEL" ] && DOCKER_ARGS="$DOCKER_ARGS -e CODE_REVIEW_MODEL"
         [ -n "$CODE_REVIEW_MAX_ITERS" ] && DOCKER_ARGS="$DOCKER_ARGS -e CODE_REVIEW_MAX_ITERS"
+        [ -n "$CODE_REVIEW_MAX_LLM_CALLS" ] && DOCKER_ARGS="$DOCKER_ARGS -e CODE_REVIEW_MAX_LLM_CALLS"
         [ -n "$CODE_REVIEW_REASONING_EFFORT" ] && DOCKER_ARGS="$DOCKER_ARGS -e CODE_REVIEW_REASONING_EFFORT"
         [ -n "$CODE_REVIEW_TEMPERATURE" ] && DOCKER_ARGS="$DOCKER_ARGS -e CODE_REVIEW_TEMPERATURE"
 
@@ -466,13 +502,15 @@ runs:
         [ -n "$SUPPLY_CHAIN_ENABLED" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_ENABLED"
         [ -n "$SUPPLY_CHAIN_MODEL" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_MODEL"
         [ -n "$SUPPLY_CHAIN_MAX_ITERS" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_MAX_ITERS"
+        [ -n "$SUPPLY_CHAIN_MAX_LLM_CALLS" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_MAX_LLM_CALLS"
         [ -n "$SUPPLY_CHAIN_REASONING_EFFORT" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_REASONING_EFFORT"
         [ -n "$SUPPLY_CHAIN_TEMPERATURE" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_TEMPERATURE"
         [ -n "$SUPPLY_CHAIN_SCAN_UNCHANGED" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUPPLY_CHAIN_SCAN_UNCHANGED"
 
         # Summary
         [ -n "$SUMMARY_ENABLED" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUMMARY_ENABLED"
         [ -n "$SUMMARY_MODEL" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUMMARY_MODEL"
+        [ -n "$SUMMARY_MAX_LLM_CALLS" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUMMARY_MAX_LLM_CALLS"
         [ -n "$SUMMARY_REASONING_EFFORT" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUMMARY_REASONING_EFFORT"
         [ -n "$SUMMARY_TEMPERATURE" ] && DOCKER_ARGS="$DOCKER_ARGS -e SUMMARY_TEMPERATURE"
 

codespy.yaml

@@ -44,6 +44,10 @@ llm:
   # Caches system prompts on the LLM provider's servers to reduce latency and costs
   enable_prompt_caching: true              # ENABLE_PROMPT_CACHING
 
+# Minimum confidence threshold for reported issues (0.0–1.0).
+# Issues below this threshold are silently discarded.
+min_confidence: 0.81                          # MIN_CONFIDENCE
+
 
 # ============================================================================
 # GIT PLATFORMS
@@ -187,6 +191,7 @@ memory:
 default_model: anthropic/claude-opus-4-6        # DEFAULT_MODEL
 extraction_model: null                          # EXTRACTION_MODEL (falls back to default_model)
 default_max_iters: 10                           # DEFAULT_MAX_ITERS
+default_max_llm_calls: 30                      # DEFAULT_MAX_LLM_CALLS
 default_reasoning_effort: medium                # DEFAULT_REASONING_EFFORT (minimal | low | medium | high)
 default_temperature: 0.2                        # DEFAULT_TEMPERATURE
 
@@ -211,6 +216,7 @@ signatures:
   supply_chain:
     enabled: true                        # SUPPLY_CHAIN_ENABLED
     max_iters: null                      # SUPPLY_CHAIN_MAX_ITERS
+    max_llm_calls: null                  # SUPPLY_CHAIN_MAX_LLM_CALLS
     model: null                          # SUPPLY_CHAIN_MODEL (Haiku 4.5)
     reasoning_effort: null               # SUPPLY_CHAIN_REASONING_EFFORT
     temperature: null                    # SUPPLY_CHAIN_TEMPERATURE
@@ -227,6 +233,7 @@ signatures:
   code_review:
     enabled: true                        # CODE_REVIEW_ENABLED
     max_iters: null                      # CODE_REVIEW_MAX_ITERS
+    max_llm_calls: null                  # CODE_REVIEW_MAX_LLM_CALLS
     model: null                          # CODE_REVIEW_MODEL
     reasoning_effort: null               # CODE_REVIEW_REASONING_EFFORT
     temperature: null                    # CODE_REVIEW_TEMPERATURE
@@ -239,6 +246,7 @@ signatures:
   # Note: doc extraction is now deterministic (no LLM) — see doc_extractor.py
   doc:
     enabled: true                        # DOC_ENABLED
+    max_llm_calls: null                  # DOC_MAX_LLM_CALLS
     model: null                          # DOC_MODEL
     reasoning_effort: null               # DOC_REASONING_EFFORT
     temperature: null                    # DOC_TEMPERATURE
@@ -251,6 +259,7 @@ signatures:
   scope:
     enabled: true                        # SCOPE_ENABLED
     max_iters: 20                        # SCOPE_MAX_ITERS
+    max_llm_calls: null                  # SCOPE_MAX_LLM_CALLS
     model: null                          # SCOPE_MODEL
     reasoning_effort: null               # SCOPE_REASONING_EFFORT
     temperature: null                    # SCOPE_TEMPERATURE
@@ -262,6 +271,7 @@ signatures:
   # Summarizer signature
   summary:
     enabled: true                        # SUMMARY_ENABLED
+    max_llm_calls: null                  # SUMMARY_MAX_LLM_CALLS
     model: null                          # SUMMARY_MODEL (falls back to default_model)
     reasoning_effort: null               # SUMMARY_REASONING_EFFORT
     temperature: null                    # SUMMARY_TEMPERATURE
@@ -273,6 +283,7 @@ signatures:
   # Auditor signature (quality assessment + recommendation after all reviews)
   audit:
     enabled: true                        # AUDIT_ENABLED
+    max_llm_calls: null                  # AUDIT_MAX_LLM_CALLS
     model: null                          # AUDIT_MODEL
     reasoning_effort: null               # AUDIT_REASONING_EFFORT
     temperature: null                    # AUDIT_TEMPERATURE

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "codespy-ai"
-version = "1.0.2"
+version = "1.0.3"
 description = "Code review agent powered by DSPy"
 readme = "README.md"
 license = "MIT"

src/codespy/__init__.py

@@ -1,3 +1,3 @@
 """codespy - Code review agent powered by DSPy."""
 
-__version__ = "1.0.2"
+__version__ = "1.0.3"

src/codespy/agents/context_safe.py

@@ -59,12 +59,22 @@ class ContextSafe(dspy.Module):
     Transparent to Hippocampus — delegates .signature to inner module.
     """
 
-    def __init__(self, module: dspy.Module, signature, tools: list | None = None, name: str = ""):
+    def __init__(
+        self,
+        module: dspy.Module,
+        signature,
+        tools: list | None = None,
+        name: str = "",
+        max_iters: int | None = None,
+        max_llm_calls: int | None = None,
+    ):
         super().__init__()
         self.module = module
         self._orig_signature = signature
         self._tools = tools
         self._name = name or signature.__name__
+        self._max_iters = max_iters
+        self._max_llm_calls = max_llm_calls
 
     @property
     def signature(self):
@@ -153,6 +163,6 @@ def _create_rlm_fallback(self) -> dspy.RLM:
         return dspy.RLM(
             self._get_current_signature(),
             tools=self._tools,
-            max_iters=10,
-            max_llm_calls=20,
+            max_iters=self._max_iters or 10,
+            max_llm_calls=self._max_llm_calls or 20,
         )

src/codespy/agents/reviewer/modules/auditor.py

@@ -146,7 +146,12 @@ def forward(
                 "NEEDS_DISCUSSION" if all_issues else "APPROVE",
             )
 
-        auditor = ContextSafe(dspy.ChainOfThought(AuditSignature), AuditSignature, name="audit")
+        auditor = ContextSafe(
+            dspy.ChainOfThought(AuditSignature),
+            AuditSignature,
+            name="audit",
+            max_llm_calls=self._settings.get_max_llm_calls("audit"),
+        )
         logger.info("Running audit...")
 
         with SignatureContext("audit", self._cost_tracker):

src/codespy/agents/reviewer/modules/code_reviewer.py

@@ -13,7 +13,6 @@
 from codespy.agents.memory.hippocampus import ContextMemory, Hippocampus
 from codespy.agents.reviewer.models import Issue, IssueCategory, ReviewContext, ScopeResult
 from codespy.agents.reviewer.modules.helpers import (
-    MIN_CONFIDENCE,
     issues_to_markdown,
     make_scope_relative,
     resolve_scope_root,
@@ -238,6 +237,8 @@ async def aforward(
                     CodeReviewSignature,
                     tools=tools,
                     name="code_review",
+                    max_iters=max_iters,
+                    max_llm_calls=self._settings.get_max_llm_calls("code_review"),
                 )
                 scoped = make_scope_relative(scope)
                 logger.info(
@@ -268,7 +269,7 @@ async def aforward(
                         )
                         issues = [
                             issue for issue in (result.issues or [])
-                            if issue.confidence >= MIN_CONFIDENCE
+                            if issue.confidence >= self._settings.min_confidence
                         ]
                         await mem.aend_episode(
                             get_memory_store(self._settings),
@@ -285,7 +286,7 @@ async def aforward(
                         )
                         issues = [
                             issue for issue in (result.issues or [])
-                            if issue.confidence >= MIN_CONFIDENCE
+                            if issue.confidence >= self._settings.min_confidence
                         ]
                 restore_repo_paths(issues, scope.subroot)
                 all_issues.extend(issues)