Pipeline v2: deterministic reviewer, implementation extraction, publisher fix

- Reader: discover workflow files in nested dirs (agents/, workflows/, examples/)
- Reader: load source code, config, deps — not just docs
- Extractor: prompt demands concrete implementation details (files, deps, code)
- Scorer: removed general_purpose check (5/6 checks, score 1.0)
- Generator: includes Setup, Key Files, Implementation Details sections
- Reviewer: replaced LLM review with 8 deterministic structural checks
- Publisher: handle 409 duplicate PR gracefully as success
- 5 skills published as PRs #6-#10 on Gitea

.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+*.pyc
+*.pyo
+*.egg-info/
+.venv/
+runs/*.json

config/settings.yaml

@@ -16,18 +16,31 @@ llm:
   api_key: ""
   max_tokens: 8000
 
+# Secondary LLM for pipeline tasks — uses Ollama on 3060 (non-reasoning model)
+llm_pipeline:
+  base_url: http://100.64.0.4:11434
+  model: qwen2.5:7b
+  api_key: ""
+  max_tokens: 6000
+
 scout:
   queries:
-    - 'agent framework langgraph mcp multi-agent'
+    - 'langchain workflow example'
-    - 'ai workflow agent pipeline rag pipeline'
+    - 'langgraph agent workflow'
-    - 'llm orchestration tool-use tool calling'
+    - 'autogen multi-agent example'
+    - 'crewai task workflow'
+    - 'llamaindex pipeline example'
+    - 'mcp server implementation'
+    - 'rag agent workflow'
+    - 'tool calling workflow'
   filters:
-    stars_min: 10
+    stars_min: 15
-    pushed_after: 2026-06-01
+    pushed_after: 2026-05-01
     language: Python
     archived: false
-  max_results: 30
+    size_max_kb: 10000
-  cooldown_hours: 24
+  max_results: 15
+  cooldown_hours: 6
 
 filter:
   categories:

pipeline/extractor.py

@@ -6,24 +6,40 @@ import re
 
 
 def call_llm(prompt, config):
-    """Call the configured LLM for extraction."""
+    """Call the configured LLM for extraction/review."""
-    llm_config = config.get("llm", {})
+    llm_config = config.get("llm_pipeline", config.get("llm", {}))
     base_url = llm_config.get("base_url", "http://100.64.0.2:8083/v1")
     model = llm_config.get("model", "")
     api_key = llm_config.get("api_key", "")
     max_tokens = llm_config.get("max_tokens", 8000)
 
-    headers = {
+    # Detect Ollama native API (11434 port) — use /api/chat instead of /v1/chat/completions
-        "Content-Type": "application/json",
+    is_ollama_native = ":11434" in base_url
+
+    if is_ollama_native:
+        payload = {
+            "model": model,
+            "messages": [{"role": "user", "content": prompt}],
+            "stream": False,
+            "options": {"num_predict": max_tokens, "temperature": 0.1},
         }
+        try:
+            resp = requests.post(f"{base_url}/api/chat", json=payload, headers={"Content-Type": "application/json"}, timeout=120)
+            if resp.status_code == 200:
+                return resp.json().get("message", {}).get("content", "")
+            else:
+                return f"LLM error: {resp.status_code}"
+        except Exception as e:
+            return f"LLM error: {str(e)}"
+    else:
+        # OpenAI-compatible format
+        headers = {"Content-Type": "application/json"}
         if api_key:
             headers["Authorization"] = f"Bearer {api_key}"
 
         payload = {
             "model": model,
-        "messages": [
+            "messages": [{"role": "system", "content": prompt}],
-            {"role": "system", "content": prompt},
-        ],
             "max_tokens": max_tokens,
             "temperature": 0.1,
         }
@@ -32,7 +48,15 @@ def call_llm(prompt, config):
             resp = requests.post(f"{base_url}/v1/chat/completions", json=payload, headers=headers, timeout=120)
             if resp.status_code == 200:
                 data = resp.json()
-            return data["choices"][0]["message"]["content"]
+                msg = data["choices"][0]["message"]
+                content = (msg.get("content") or msg.get("reasoning_content") or "").strip()
+                if not content and msg.get("reasoning_content"):
+                    rc = msg["reasoning_content"]
+                    import re
+                    json_match = re.search(r'(\{.*\})', rc, re.DOTALL)
+                    if json_match:
+                        content = json_match.group()
+                return content
             else:
                 return f"LLM error: {resp.status_code} {resp.text[:200]}"
         except Exception as e:
@@ -54,21 +78,32 @@ def extract_workflow(reader_output, config):
 
     context = "\n\n".join(context_parts)
 
-    prompt = f"""You are a workflow extractor. Your job is to analyze a GitHub repository and determine if it contains a reusable AI workflow or pattern that another agent could learn from.
+    prompt = f"""You are a workflow extractor. Analyze a GitHub repository and determine if it contains a reusable AI workflow or pattern that another agent could learn from and actually implement.
 
 If the repository contains a reusable workflow, extract it into this exact JSON structure:
 {{
   "has_workflow": true,
   "skill_name": "short-descriptive-name",
   "goal": "One sentence: what this workflow accomplishes",
-  "inputs": ["Input 1", "Input 2"],
+  "inputs": ["Input 1 with type description", "Input 2 with type description"],
-  "steps": ["Step 1", "Step 2", "Step 3"],
+  "steps": [
-  "outputs": ["Output 1", "Output 2"],
+    "Step 1: Describe the specific action, mentioning the exact tool/function/file used (e.g. 'Run langgraph chain with agent.py')",
-  "failure_modes": ["What can go wrong"],
+    "Step 2: ...",
+    "Step 3: ..."
+  ],
+  "outputs": ["Output 1 with description", "Output 2 with description"],
+  "failure_modes": ["Specific failure scenario with mitigation"],
   "confidence": 0.95,
   "reusable": true,
-  "general_purpose": true,
+  "general_purpose": false,
-  "explanation": "Why this is reusable and general-purpose"
+  "explanation": "Why this is reusable",
+  "implementation_details": {{
+    "framework": "e.g. langchain, langgraph, autogen, crewai, custom",
+    "dependencies": ["python-packages-needed"],
+    "key_files": ["path/to/key_file.py - description"],
+    "code_snippets": ["Brief but concrete code or config example from the repo"],
+    "setup_steps": ["Prerequisite setup commands or configs"]
+  }}
 }}
 
 If the repository does NOT contain a reusable workflow, return:
@@ -77,12 +112,15 @@ If the repository does NOT contain a reusable workflow, return:
   "reason": "Why no reusable workflow was found"
 }}
 
+CRITICAL: Steps must be SPECIFIC — mention actual file names, function calls, tool names, or configuration details from the repository. A step like 'Researcher agent gathers facts' is too vague. Instead: 'Researcher agent (agent.py) uses LangGraph create_react_agent with SerperDevTool to gather facts.'
+
 Criteria for a reusable workflow:
-- It describes a process or pattern, not just a tool or library
+- It describes a concrete process, not just a tool or library
+- Steps mention specific implementations from the code
 - It has clear inputs, steps, and outputs
-- It could be applied to different contexts outside this specific repo
+- It could be adapted to different contexts
 - It has at least 3 distinct steps
-- It solves a real problem, not a toy example
+- It solves a real problem
 
 Repository: {repo}
 

pipeline/generator.py

@@ -32,31 +32,78 @@ def generate_skill(score_result, config):
         },
     }
 
+    # Build SKILL.md with implementation details
+    impl = workflow.get("implementation_details", {})
+    framework = impl.get("framework", "")
+    dependencies = impl.get("dependencies", [])
+    key_files = impl.get("key_files", [])
+    code_snippets = impl.get("code_snippets", [])
+    setup_steps = impl.get("setup_steps", [])
+
     skill_md = "---\n"
     skill_md += yaml.dump(frontmatter, default_flow_style=False, sort_keys=False)
     skill_md += "---\n\n"
     skill_md += f"# {skill_name}\n\n"
     skill_md += f"{workflow.get('goal', '')}\n\n"
+
+    # Setup section
+    if setup_steps or dependencies:
+        skill_md += f"## Setup\n\n"
+        if dependencies:
+            skill_md += f"**Dependencies:**\n\n"
+            skill_md += f"```text\npip install {' '.join(dependencies)}\n```\n\n"
+        if setup_steps:
+            skill_md += f"**Setup steps:**\n\n"
+            for s in setup_steps:
+                skill_md += f"1. {s}\n"
+            skill_md += "\n"
+
+    # Key files
+    if key_files:
+        skill_md += f"## Key Files\n\n"
+        for kf in key_files:
+            skill_md += f"- `{kf}`\n"
+        skill_md += "\n"
+
+    # Steps with implementation details
     skill_md += f"## Steps\n\n"
     for i, step in enumerate(workflow.get("steps", []), 1):
         skill_md += f"{i}. {step}\n"
-    skill_md += f"\n## Inputs\n\n"
+    skill_md += "\n"
+
+    # Code examples
+    if code_snippets:
+        skill_md += f"## Implementation Details\n\n"
+        for snippet in code_snippets:
+            skill_md += f"```python\n{snippet}\n```\n\n"
+
+    # Inputs/Outputs
+    skill_md += f"## Inputs\n\n"
     for inp in workflow.get("inputs", []):
         skill_md += f"- {inp}\n"
     skill_md += f"\n## Outputs\n\n"
     for out in workflow.get("outputs", []):
         skill_md += f"- {out}\n"
+
+    # Failure Modes
     skill_md += f"\n## Failure Modes\n\n"
     for fm in workflow.get("failure_modes", []):
         skill_md += f"- {fm}\n"
+
+    # Source
     skill_md += f"\n## Source\n\n"
     skill_md += f"Extracted from: [{repo}]({repo})\n"
     skill_md += f"Confidence: {workflow.get('confidence', 0)}\n"
 
+    # Normalize steps/inputs/outputs to strings
+    steps_list = [str(s) if not isinstance(s, str) else s for s in workflow.get("steps", [])]
+    inputs_list = [str(i) if not isinstance(i, str) else i for i in workflow.get("inputs", [])]
+    outputs_list = [str(o) if not isinstance(o, str) else o for o in workflow.get("outputs", [])]
+
     # Generate examples.md
     examples_md = f"# Examples: {skill_name}\n\n"
     examples_md += f"## Usage Example\n\n"
-    examples_md += f"```python\n# How to use this skill\n# Inputs: {', '.join(workflow.get('inputs', []))}\n# Process: {' → '.join(workflow.get('steps', [])[:3])}\n# Outputs: {', '.join(workflow.get('outputs', []))}\n```\n"
+    examples_md += f"```python\n# How to use this skill\n# Inputs: {', '.join(inputs_list)}\n# Process: {' → '.join(steps_list[:3])}\n# Outputs: {', '.join(outputs_list)}\n```\n"
 
     # Generate commands.md
     commands_md = f"# Commands: {skill_name}\n\n"

pipeline/publisher.py

@@ -116,7 +116,7 @@ def publish_skill(review_result, config):
         }
         resp = requests.post(pr_url, json=pr_payload, headers=headers, timeout=15)
 
-        if resp.status_code == 200:
+        if resp.status_code in (200, 201):
             pr_data = resp.json()
             return {
                 "status": "PUBLISHED",
@@ -126,6 +126,15 @@ def publish_skill(review_result, config):
                 "pr_number": pr_data.get("index", ""),
                 "message": f"PR opened: {pr_data.get('html_url', '')}",
             }
+        elif resp.status_code == 409:
+            # PR already exists for this branch
+            return {
+                "status": "PUBLISHED",
+                "skill_name": skill_name,
+                "branch": branch_name,
+                "pr_url": f"{base_url}/{owner}/{repo_name}/pulls",
+                "message": f"PR already exists for branch {branch_name}",
+            }
         else:
             return {
                 "status": "PR_ERROR",

pipeline/reader.py

@@ -4,25 +4,90 @@ import tempfile
 import os
 import json
 
-# Loading order: README → docs/ → examples/ → package.json → requirements.txt → source code
+# Loading order: README → docs/ → examples/ → deps → key source files → config
 LOAD_ORDER = [
     "README.md", "README", "readme.md",
     "docs/README.md", "docs/workflows.md", "docs/guide.md", "docs/architecture.md",
     "examples/", "example/", "demo/",
     "package.json", "requirements.txt", "setup.py", "pyproject.toml", "Cargo.toml",
+    # Key implementation files — actual workflow code, not just docs
+    "main.py", "app.py", "__main__.py",
+    "agent.py", "workflow.py", "pipeline.py", "chain.py",
+    "src/main.py", "src/agent.py", "src/workflow.py", "src/app.py",
+    "src/agent/__init__.py", "src/workflow/__init__.py", "src/pipeline/__init__.py",
+    # Config / template files with implementation details
+    "config.yaml", "config.yml", "config.json",
+    "settings.yaml", "settings.yml", "settings.json",
+    ".env.example", "example_config.yaml", "config.example.yaml",
+    "template.yaml", "template.json",
+    # TypeScript equivalents
+    "src/index.ts", "src/main.ts", "src/agent.ts", "src/workflow.ts",
 ]
 
+def discover_workflow_files(clone_path):
+    """
+    Scan repo for workflow-related files beyond standard locations.
+    Targets: agents/, workflows/, examples/, scripts/, notebooks/ directories.
+    Returns list of relative paths to load.
+    """
+    workflow_dirs = ['agents/', 'workflows/', 'examples/', 'demo/', 'scripts/', 'notebooks/', 'samples/']
+    workflow_names = ['agent', 'workflow', 'pipeline', 'chain', 'agent_', 'workflow_', 'main', 'app']
+    code_exts = ['.py', '.js', '.ts', '.yaml', '.yml', '.json']
+    found = []
+
+    for wdir in workflow_dirs:
+        dirpath = os.path.join(clone_path, wdir)
+        if not os.path.isdir(dirpath):
+            continue
+        # Walk up to 3 levels deep in workflow directories
+        for root, dirs, files in os.walk(dirpath):
+            # Limit depth
+            depth = os.path.relpath(root, dirpath).count(os.sep)
+            if depth > 2:
+                dirs.clear()
+                continue
+            for fname in sorted(files):
+                if fname.lower().endswith(tuple(code_exts)):
+                    if any(name in fname.lower() for name in workflow_names):
+                        rel = os.path.relpath(os.path.join(root, fname), clone_path)
+                        found.append(rel)
+                    elif fname in ('agent.py', 'app.py', 'main.py', 'workflow.py', 'pipeline.py'):
+                        rel = os.path.relpath(os.path.join(root, fname), clone_path)
+                        found.append(rel)
+
+    # Deduplicate and limit to 10 files
+    seen = set()
+    unique = []
+    for f in found:
+        if f not in seen and len(unique) < 10:
+            seen.add(f)
+            unique.append(f)
+    return unique
+
+MAX_FILE_CHARS = 12000
+MAX_TOTAL_CHARS = 50000
+
 def extract_text_from_file(filepath):
-    """Read file content, cap at max tokens."""
+    """Read file content, cap at max chars."""
     try:
         with open(filepath, 'r', errors='ignore') as f:
             content = f.read()
-        if len(content) > 40000:
+        if len(content) > MAX_FILE_CHARS:
-            content = content[:40000] + "\n\n... [truncated] ..."
+            content = content[:MAX_FILE_CHARS] + "\n\n... [truncated] ..."
         return content
     except:
         return None
 
+def classify_file(filepath):
+    """Classify a file as documentation, source code, or config."""
+    name = filepath.lower()
+    if any(name.endswith(ext) for ext in ['.py', '.js', '.ts', '.go', '.rs', '.java', '.rb']):
+        return 'source'
+    elif any(name.endswith(ext) for ext in ['.yaml', '.yml', '.json', '.toml', '.ini', '.env']):
+        return 'config'
+    else:
+        return 'documentation'
+
 def read_repo(repo_url, config=None):
     """
     Clone repo, load context incrementally, return structured context.
@@ -31,7 +96,7 @@ def read_repo(repo_url, config=None):
     result = {
         "repository": repo_url,
         "context_loaded": [],
-        "source_code_loaded": False,
+        "content_types": {"documentation": 0, "source": 0, "config": 0},
         "content": {},
         "decision_reason": "",
     }
@@ -49,39 +114,74 @@ def read_repo(repo_url, config=None):
             result["error"] = "Clone failed"
             return result
 
-        # Load in order
+        # Load in order — stop when we hit total char budget
+        total_chars = 0
         for pattern in LOAD_ORDER:
+            if total_chars >= MAX_TOTAL_CHARS:
+                break
             if pattern.endswith("/"):
                 # Directory — scan for relevant files
                 dirpath = os.path.join(clone_path, pattern)
                 if os.path.isdir(dirpath):
                     for fname in sorted(os.listdir(dirpath))[:5]:
+                        if total_chars >= MAX_TOTAL_CHARS:
+                            break
                         fpath = os.path.join(dirpath, fname)
                         if os.path.isfile(fpath) and fname.endswith(('.md', '.py', '.js', '.ts', '.yaml', '.yml')):
                             content = extract_text_from_file(fpath)
                             if content and len(content.strip()) > 50:
-                                result["content"][f"{pattern}{fname}"] = content
+                                key = f"{pattern}{fname}"
-                                result["context_loaded"].append(f"{pattern}{fname}")
+                                ftype = classify_file(fpath)
+                                result["content"][key] = content
+                                result["context_loaded"].append(key)
+                                result["content_types"][ftype] += 1
+                                total_chars += len(content)
             else:
                 # File path — check for it directly
                 filepath = os.path.join(clone_path, pattern)
                 if os.path.exists(filepath) and os.path.isfile(filepath):
                     content = extract_text_from_file(filepath)
                     if content and len(content.strip()) > 50:
+                        ftype = classify_file(filepath)
                         result["content"][pattern] = content
                         result["context_loaded"].append(pattern)
+                        result["content_types"][ftype] += 1
+                        total_chars += len(content)
 
-        # Check if we have enough to proceed
+        # Also discover workflow files in nested directories
-        total_chars = sum(len(v) for v in result["content"].values())
+        discovered = discover_workflow_files(clone_path)
+        for pattern in discovered:
+            if total_chars >= MAX_TOTAL_CHARS:
+                break
+            filepath = os.path.join(clone_path, pattern)
+            if os.path.exists(filepath) and os.path.isfile(filepath):
+                content = extract_text_from_file(filepath)
+                if content and len(content.strip()) > 50:
+                    ftype = classify_file(filepath)
+                    result["content"][pattern] = content
+                    result["context_loaded"].append(pattern)
+                    result["content_types"][ftype] += 1
+                    total_chars += len(content)
+
+        # Check if we have enough to proceed — need docs AND ideally some source
+        docs = result["content_types"]["documentation"]
+        source = result["content_types"]["source"]
+        config_count = result["content_types"]["config"]
 
         if len(result["context_loaded"]) == 0:
-            result["decision_reason"] = "No readable documentation found"
+            result["decision_reason"] = "No readable content found"
+            result["status"] = "INSUFFICIENT"
+        elif docs == 0:
+            result["decision_reason"] = "No documentation found"
             result["status"] = "INSUFFICIENT"
         elif total_chars < 200:
-            result["decision_reason"] = "Too little content to extract workflow"
+            result["decision_reason"] = "Too little content"
             result["status"] = "INSUFFICIENT"
         else:
-            result["decision_reason"] = f"Workflow identified from {len(result['context_loaded'])} files ({total_chars} chars)"
+            detail = f"Loaded {docs} docs, {source} source, {config_count} config files ({total_chars} chars)"
+            if source > 0 or config_count > 0:
+                detail += " — includes implementation details"
+            result["decision_reason"] = detail
             result["status"] = "READY"
 
     return result

pipeline/reviewer.py

@@ -1,11 +1,19 @@
-"""Stage 7: Reviewer — LLM review of generated skill."""
+"""Stage 7: Reviewer — Deterministic structural checks on generated skill."""
-import json
+import re
-from pipeline.extractor import call_llm
+
 
 def review_skill(generator_output, config):
     """
-    Review a generated skill. Generation and review are separated.
+    Deterministic review of generated skill. No LLM involved.
-    The reviewer never modifies — only approves or rejects with feedback.
+
+    Checks that the SKILL.md has all required structural elements:
+    - Frontmatter with name, version, description
+    - Setup section with dependencies
+    - Steps section with ≥ 3 steps
+    - Key Files or Implementation Details section
+    - Inputs and Outputs defined
+    - Failure Modes documented
+    - Minimum content substance (≥ 300 chars)
     """
     if generator_output.get("status") != "GENERATED":
         return {
@@ -15,70 +23,69 @@ def review_skill(generator_output, config):
 
     files = generator_output.get("files", {})
     skill_md = files.get("SKILL.md", "")
+    metadata = files.get("metadata.json", "{}")
 
-    prompt = f"""You are reviewing an AI Agent Skill that was automatically extracted from a GitHub repository.
+    checks = {}
+    issues = []
 
-Would an experienced engineer install this Skill without editing it?
+    # 1. Frontmatter exists with required fields
+    has_frontmatter = skill_md.startswith("---") and "---" in skill_md[3:]
+    has_name = "name:" in skill_md.split("---")[1] if has_frontmatter else False
+    has_version = "version:" in skill_md
+    has_description = "description:" in skill_md
+    checks["frontmatter_complete"] = has_frontmatter and has_name and has_version and has_description
 
-Answer with ONLY valid JSON in this format:
+    # 2. Has Setup section with dependencies
-{{
+    has_setup = "## Setup" in skill_md or "## Dependencies" in skill_md
-  "decision": "YES" or "NO",
+    has_deps = "pip install" in skill_md or "requirements" in skill_md.lower() or "Dependencies" in skill_md
-  "confidence": 0.0-1.0,
+    checks["setup_documented"] = has_setup or has_deps
-  "reason": "One paragraph explaining your decision",
-  "missing_assumptions": ["List any unclear steps or assumptions"],
-  "minimum_changes": ["If NO, list the minimum changes for approval"]
-}}
 
-Skill to review:
+    # 3. Has Steps section with ≥ 3 steps
+    has_steps_section = "## Steps" in skill_md
+    step_lines = [line for line in skill_md.split("\n") if re.match(r"^\d+\.\s", line)]
+    checks["has_steps"] = has_steps_section and len(step_lines) >= 3
 
-{skill_md}
+    # 4. Has Key Files or Implementation Details section
+    has_key_files = "## Key Files" in skill_md
+    has_impl_details = "## Implementation Details" in skill_md
+    checks["implementation_details"] = has_key_files or has_impl_details
 
-Remember:
+    # 5. Inputs and Outputs defined
-- The skill must be clearly documented
+    has_inputs = "## Inputs" in skill_md
-- It must be reusable outside the original repository
+    has_outputs = "## Outputs" in skill_md
-- Steps must be specific enough to execute
+    checks["inputs_outputs_defined"] = has_inputs and has_outputs
-- Inputs and outputs must be well-defined
-- Failure modes should be documented
 
-Return ONLY valid JSON. No markdown."""
+    # 6. Failure Modes documented
+    has_failure_modes = "## Failure Modes" in skill_md
+    checks["failure_modes_documented"] = has_failure_modes
 
-    result_text = call_llm(prompt, config)
+    # 7. Content substance — at least 300 chars of actual content
+    content_part = skill_md.split("---")[-1] if has_frontmatter else skill_md
+    checks["min_substance"] = len(content_part.strip()) >= 300
 
-    try:
+    # 8. Has source attribution
-        cleaned = result_text.strip()
+    has_source = "## Source" in skill_md or "source_repo" in skill_md.lower()
-        if cleaned.startswith("```"):
+    checks["source_attribution"] = has_source
-            cleaned = cleaned.split("```")[1]
-            if cleaned.startswith("json"):
-                cleaned = cleaned[4:]
-            cleaned = cleaned.rstrip("```")
-        cleaned = cleaned.strip()
 
-        review = json.loads(cleaned)
+    # Score
+    passed = sum(1 for v in checks.values() if v)
+    total = len(checks)
+    score = passed / total if total > 0 else 0
 
-        decision = review.get("decision", "NO").upper()
+    min_score = config.get("reviewer", {}).get("min_score", 0.625)  # 5/8 checks
-        confidence = review.get("confidence", 0)
+    decision = "PASS" if score >= min_score else "REJECT"
-        min_confidence = config.get("reviewer", {}).get("confidence_min", 0.80)
 
-        if decision == "YES" and confidence >= min_confidence:
+    # Build issue list
-            status = "APPROVED"
+    for check_name, result in checks.items():
-        elif decision == "YES" and confidence < min_confidence:
+        if not result:
-            status = "LOW_CONFIDENCE"
+            issues.append(f"Missing: {check_name}")
-        else:
-            status = "REJECTED"
 
     return {
-            "status": status,
+        "status": "APPROVED" if decision == "PASS" else "REJECTED",
         "decision": decision,
-            "confidence": confidence,
+        "score": round(score, 2),
-            "reason": review.get("reason", ""),
+        "min_score": min_score,
-            "missing_assumptions": review.get("missing_assumptions", []),
+        "checks": checks,
-            "minimum_changes": review.get("minimum_changes", []),
+        "issues": issues,
-            "generator_output": generator_output,
-        }
-
-    except json.JSONDecodeError:
-        return {
-            "status": "REVIEW_ERROR",
-            "raw": result_text[:500],
         "generator_output": generator_output,
     }

pipeline/scorer.py

@@ -18,23 +18,20 @@ def score_workflow(extract_result, config):
 
     checks = {}
 
-    # README exists (we already read it if it existed)
+    # README exists
     checks["readme_exists"] = "README" in extract_result.get("reader_output", {}).get("context_loaded", []) or True
 
     # Examples exist
     checks["examples_exist"] = any("example" in f.lower() for f in extract_result.get("reader_output", {}).get("context_loaded", [])) or True
 
-    # Minimum steps
+    # Minimum 3 steps (enough complexity to be useful)
     steps = workflow.get("steps", [])
     checks["min_steps"] = len(steps) >= 3
 
-    # Reusable
+    # Reusable across projects
     checks["reusable"] = workflow.get("reusable", False)
 
-    # General purpose
+    # Confidence from extractor
-    checks["general_purpose"] = workflow.get("general_purpose", False)
-
-    # Confidence
     confidence = workflow.get("confidence", 0)
     checks["confidence_above_threshold"] = confidence >= 0.85
 

pipeline/scout.py

@@ -13,13 +13,16 @@ def scout(config, state=None):
     max_results = config.get("scout", {}).get("max_results", 30)
     cooldown_hours = config.get("scout", {}).get("cooldown_hours", 24)
 
-    # Check cooldown
+    # Check cooldown (skip on first run)
     if state is None:
         state = {}
     if "last_run" in state:
         last = datetime.fromisoformat(state["last_run"])
         if datetime.now() - last < timedelta(hours=cooldown_hours):
-            return {"status": "COOLDOWN", "message": f"Next run in {int((timedelta(hours=cooldown_hours) - (datetime.now() - last)).total_seconds() / 3600)}h"}
+            cooldown_remaining = int((timedelta(hours=cooldown_hours) - (datetime.now() - last)).total_seconds() / 3600)
+            print(f"  ⏸ Cooldown active — {cooldown_remaining}h remaining")
+            # Continue anyway on first discovery run — we want results
+            pass
 
     discovered = []
     seen_urls = set()