fix: repository analysis caching and PR creation 404 issues

src/agents/repository_analysis_agent/agent.py

@@ -38,6 +38,10 @@ async def execute(self, **kwargs) -> AgentResult:
             await analyze_pr_history(state, request.max_prs)
             await analyze_contributing_guidelines(state)
 
+            # Only generate recommendations if we have basic repository data
+            if not state.repository_features.language:
+                raise ValueError("Unable to determine repository language - cannot generate appropriate rules")
+
             state.recommendations = _default_recommendations(state)
             validate_recommendations(state)
             response = summarize_analysis(state, request)

src/agents/repository_analysis_agent/nodes.py

@@ -8,6 +8,7 @@
 
 from __future__ import annotations
 
+import logging
 import textwrap
 from typing import Any
 
@@ -32,6 +33,9 @@ async def analyze_repository_structure(state: RepositoryAnalysisState) -> None:
     installation_id = state.installation_id
 
     repo_data = await github_client.get_repository(repo, installation_id=installation_id)
+    if not repo_data:
+        raise ValueError(f"Could not fetch repository data for {repo}")
+
     workflows = await github_client.list_directory_any_auth(
         repo_full_name=repo, path=".github/workflows", installation_id=installation_id
     )
@@ -42,7 +46,7 @@ async def analyze_repository_structure(state: RepositoryAnalysisState) -> None:
         has_codeowners=bool(await github_client.get_file_content(repo, ".github/CODEOWNERS", installation_id)),
         has_workflows=bool(workflows),
         workflow_count=len(workflows or []),
-        language=(repo_data or {}).get("language"),
+        language=repo_data.get("language"),
         contributor_count=len(contributors),
         pr_count=0,
     )
@@ -54,8 +58,14 @@ async def analyze_pr_history(state: RepositoryAnalysisState, max_prs: int) -> No
     installation_id = state.installation_id
     prs = await github_client.list_pull_requests(repo, installation_id=installation_id, state="all", per_page=max_prs)
 
+    if prs is None:
+        # If PR listing fails, continue with empty samples rather than failing
+        state.pr_samples = []
+        state.repository_features.pr_count = 0
+        return
+
     samples: list[PullRequestSample] = []
-    for pr in prs or []:
+    for pr in prs:
         samples.append(
             PullRequestSample(
                 number=pr.get("number", 0),
@@ -215,19 +225,28 @@ def _default_recommendations(
 
     Currently, validators like `author_team_is` and `file_patterns` operate independently.
     """
+    logger = logging.getLogger(__name__)
+
     recommendations: list[RuleRecommendation] = []
 
     # Get language-specific patterns based on repository analysis
-    source_patterns, test_patterns = _get_language_specific_patterns(state.repository_features.language)
+    language = state.repository_features.language
+    source_patterns, test_patterns = _get_language_specific_patterns(language)
+
+    logger.info(
+        f"Generating recommendations for {state.repository_full_name}: language={language}, pr_count={state.repository_features.pr_count}"
+    )
 
     # Analyze PR history for bad habits
     pr_issues = _analyze_pr_bad_habits(state)
 
     # Require tests when source code changes.
     # This is especially important if we detect missing tests in PR history
-    test_reasoning = f"Default guardrail for code changes without tests. Patterns adapted for {state.repository_features.language or 'multi-language'} repository."
+    test_reasoning = f"Repository analysis for {state.repository_full_name}. Language: {language or 'unknown'}. Patterns adapted for {language or 'multi-language'} repository."
     if pr_issues.get("missing_tests", 0) > 0:
         test_reasoning += f" Detected {pr_issues['missing_tests']} recent PRs without test files."
+    if state.contributing_analysis.content and state.contributing_analysis.requires_tests:
+        test_reasoning += " Contributing guidelines explicitly require tests."
 
     # Build YAML rule with proper indentation
     # parameters: is at column 0, source_patterns: at column 2, list items at column 4
@@ -239,55 +258,84 @@ def _default_recommendations(
 severity: medium
 event_types:
   - pull_request
-parameters:
+    parameters:
   source_patterns:
 {source_patterns_yaml}
   test_patterns:
 {test_patterns_yaml}
 """
 
+    confidence = 0.74
+    if pr_issues.get("missing_tests", 0) > 0:
+        confidence = 0.85
+    if state.contributing_analysis.content and state.contributing_analysis.requires_tests:
+        confidence = min(0.95, confidence + 0.1)
+
     recommendations.append(
         RuleRecommendation(
             yaml_rule=yaml_content.strip(),
-            confidence=0.74 if pr_issues.get("missing_tests", 0) == 0 else 0.85,
+            confidence=confidence,
             reasoning=test_reasoning,
             strategy_used="hybrid",
         )
     )
 
     # Require description in PR body.
     # Increase confidence if we detect short titles in PR history (indicator of missing context)
-    desc_reasoning = "Encourage context for reviewers; lightweight default."
+    desc_reasoning = f"Repository analysis for {state.repository_full_name}."
     if pr_issues.get("short_titles", 0) > 0:
         desc_reasoning += f" Detected {pr_issues['short_titles']} PRs with very short titles (likely missing context)."
+    else:
+        desc_reasoning += " Encourages context for reviewers; lightweight default."
+
+    desc_confidence = 0.68
+    if pr_issues.get("short_titles", 0) > 0:
+        desc_confidence = 0.80
 
     recommendations.append(
         RuleRecommendation(
             yaml_rule=textwrap.dedent(
                 """
                 description: "Ensure PRs include context"
-                enabled: true
+enabled: true
                 severity: low
-                event_types:
-                  - pull_request
-                parameters:
+event_types:
+  - pull_request
+    parameters:
                   min_description_length: 50
                 """
             ).strip(),
-            confidence=0.68 if pr_issues.get("short_titles", 0) == 0 else 0.80,
+            confidence=desc_confidence,
             reasoning=desc_reasoning,
             strategy_used="static",
         )
     )
 
-    # If contributing guidelines require tests, increase confidence
-    if state.contributing_analysis.content is not None and state.contributing_analysis.requires_tests:
-        # Find the test rule and boost its confidence
-        for rec in recommendations:
-            if "tests" in rec.yaml_rule.lower():
-                rec.confidence = min(0.95, rec.confidence + 0.1)
-                rec.reasoning += " Contributing guidelines explicitly require tests."
+    # Add a repository-specific rule if we detect specific patterns
+    if state.repository_features.has_workflows:
+        workflow_rule = textwrap.dedent(
+            """
+            description: "Protect CI/CD workflows"
+enabled: true
+            severity: high
+event_types:
+  - pull_request
+    parameters:
+              file_patterns:
+                - ".github/workflows/**"
+            """
+        ).strip()
+
+        recommendations.append(
+            RuleRecommendation(
+                yaml_rule=workflow_rule,
+                confidence=0.90,
+                reasoning=f"Repository {state.repository_full_name} has {state.repository_features.workflow_count} workflows that should be protected.",
+                strategy_used="static",
+            )
+        )
 
+    logger.info(f"Generated {len(recommendations)} recommendations for {state.repository_full_name}")
     return recommendations
 
 

src/agents/repository_analysis_agent/prompts.py

@@ -83,14 +83,8 @@
 severity: "medium"
 event_types:
   - pull_request
-conditions:
-  - type: "condition_type"
-    parameters:
-      key: "value"
-actions:
-  - type: "action_type"
-    parameters:
-      key: "value"
+parameters:
+  key: "value"
 ```
 
 Make sure the rule is functional and follows best practices.

src/api/recommendations.py

@@ -48,7 +48,9 @@ async def recommend_rules(
         if not request.repository_full_name or "/" not in request.repository_full_name:
             raise HTTPException(status_code=400, detail="Invalid repository name format. Expected 'owner/repo'")
 
-        cache_key = f"repo_analysis:{request.repository_full_name}"
+        # Include authentication context in cache key to ensure different access levels get different results
+        auth_context = request.installation_id or request.user_token or "anonymous"
+        cache_key = f"repo_analysis:{request.repository_full_name}:{auth_context}"
         cached_result = await get_cache(cache_key)
 
         if cached_result:
@@ -57,6 +59,7 @@ async def recommend_rules(
                 "cache_hit",
                 operation="repository_analysis",
                 subject_ids=[request.repository_full_name],
+                auth_context=auth_context,
                 cached=True,
             )
             return RepositoryAnalysisResponse(**cached_result)
@@ -85,6 +88,8 @@ async def recommend_rules(
                 decision="failed",
                 error=result.message,
             )
+            # Clear any cached results for this repository to ensure fresh analysis on retry
+            await set_cache(cache_key, None, ttl=1)  # Use 1 second TTL to effectively clear cache
             raise HTTPException(status_code=500, detail=result.message)
 
         analysis_response = result.data.get("analysis_response")
@@ -161,6 +166,18 @@ async def proceed_with_pr(request: ProceedWithPullRequestRequest) -> ProceedWith
             branch=request.branch_name,
             existing_sha=existing_branch_sha,
         )
+        # Verify the branch points to the correct base
+        if existing_branch_sha != base_sha:
+            log_structured(
+                logger,
+                "branch_sha_mismatch",
+                operation="proceed_with_pr",
+                subject_ids=[repo],
+                branch=request.branch_name,
+                existing_sha=existing_branch_sha,
+                expected_sha=base_sha,
+                warning="Branch exists but points to different SHA than base branch",
+            )
     else:
         # Create new branch
         created_ref = await github_client.create_git_ref(repo, request.branch_name, base_sha, **auth_ctx)
@@ -182,6 +199,15 @@ async def proceed_with_pr(request: ProceedWithPullRequestRequest) -> ProceedWith
                     "The branch may already exist or you may not have permission to create branches."
                 ),
             )
+        log_structured(
+            logger,
+            "branch_created",
+            operation="proceed_with_pr",
+            subject_ids=[repo],
+            branch=request.branch_name,
+            base_branch=base_branch,
+            new_sha=created_ref.get("object", {}).get("sha"),
+        )
 
     file_result = await github_client.create_or_update_file(
         repo_full_name=repo,
@@ -209,6 +235,17 @@ async def proceed_with_pr(request: ProceedWithPullRequestRequest) -> ProceedWith
             ),
         )
 
+    commit_sha = (file_result.get("commit") or {}).get("sha")
+    log_structured(
+        logger,
+        "file_created",
+        operation="proceed_with_pr",
+        subject_ids=[repo],
+        branch=request.branch_name,
+        file_path=request.file_path,
+        commit_sha=commit_sha,
+    )
+
     pr = await github_client.create_pull_request(
         repo_full_name=repo,
         title=request.pr_title,
@@ -237,16 +274,62 @@ async def proceed_with_pr(request: ProceedWithPullRequestRequest) -> ProceedWith
         )
 
     pr_url = pr.get("html_url", "")
-    if not pr_url:
+    pr_number = pr.get("number")
+    if not pr_url or not pr_number:
         log_structured(
             logger,
-            "pr_url_missing",
+            "pr_creation_incomplete",
             operation="proceed_with_pr",
             subject_ids=[repo],
             pr_data=pr,
-            error="PR created but html_url is missing",
+            pr_url=pr_url,
+            pr_number=pr_number,
+            error="PR creation response missing required fields",
+        )
+        raise HTTPException(status_code=500, detail="PR was created but response is incomplete")
+
+    # Validate the PR URL is a proper GitHub URL format
+    if not pr_url.startswith("https://github.com/") or "/pull/" not in pr_url:
+        log_structured(
+            logger,
+            "pr_url_invalid",
+            operation="proceed_with_pr",
+            subject_ids=[repo],
+            pr_url=pr_url,
+            pr_number=pr_number,
+            error="PR URL is not a valid GitHub pull request URL",
+        )
+        raise HTTPException(status_code=500, detail="PR was created but returned invalid URL format")
+
+    # Validate PR number is reasonable
+    if not isinstance(pr_number, int) or pr_number <= 0:
+        log_structured(
+            logger,
+            "pr_number_invalid",
+            operation="proceed_with_pr",
+            subject_ids=[repo],
+            pr_url=pr_url,
+            pr_number=pr_number,
+            error="PR number is invalid",
+        )
+        raise HTTPException(status_code=500, detail="PR was created but returned invalid PR number")
+
+    # Double-check URL format one more time
+    expected_url_pattern = f"https://github.com/{repo}/pull/{pr_number}"
+    if pr_url != expected_url_pattern:
+        log_structured(
+            logger,
+            "pr_url_mismatch",
+            operation="proceed_with_pr",
+            subject_ids=[repo],
+            expected_url=expected_url_pattern,
+            actual_url=pr_url,
+            pr_number=pr_number,
+            error="PR URL doesn't match expected pattern",
+        )
+        raise HTTPException(
+            status_code=500, detail=f"PR URL mismatch: expected {expected_url_pattern} but got {pr_url}"
         )
-        raise HTTPException(status_code=500, detail="PR was created but URL is missing")
 
     log_structured(
         logger,
@@ -255,11 +338,12 @@ async def proceed_with_pr(request: ProceedWithPullRequestRequest) -> ProceedWith
         subject_ids=[repo],
         decision="success",
         branch=request.branch_name,
-        pr_number=pr.get("number"),
+        pr_number=pr_number,
+        pr_url=pr_url,
     )
 
     return ProceedWithPullRequestResponse(
-        pull_request_url=pr.get("html_url", ""),
+        pull_request_url=pr_url,
         branch_name=request.branch_name,
         base_branch=base_branch,
         file_path=request.file_path,