diff --git a/app/cli/core/inspector.py b/app/cli/core/inspector.py
index 963071d..ff0d2f6 100644
--- a/app/cli/core/inspector.py
+++ b/app/cli/core/inspector.py
@@ -9,7 +9,7 @@
 # Double-brace every dict/format literal inside the script so .format() ignores them.
 # Only {path} is a real substitution placeholder.
 _INSPECT_SCRIPT = """\
-import os, json, sys
+import os, json, sys, traceback as _tb
 
 path = {path!r}
 ext = os.path.splitext(path)[1].lower()
@@ -45,211 +45,226 @@
     print(json.dumps(result))
     sys.exit(0)
 
-# ── pickle-based artifacts ──────────────────────────────────────────────────
-import pickle
-
-with open(path, "rb") as f:
-    obj = pickle.load(f)
-
-module = type(obj).__module__ or ""
-class_name = type(obj).__name__
-
-# ── framework detection (order matters) ────────────────────────────────────
-framework = "generic"
-extra = {{}}
-
-# sentence-transformers (check before transformers — it wraps HF models)
+# ── pickle-based artifacts ──────────────────────────────────────────
 try:
-    from sentence_transformers import SentenceTransformer
-    if isinstance(obj, SentenceTransformer):
-        framework = "sentence_transformers"
-except Exception:
-    pass
-
-# transformers PreTrainedModel
-if framework == "generic":
-    try:
-        from transformers import PreTrainedModel
-        if isinstance(obj, PreTrainedModel):
-            framework = "transformers"
-    except Exception:
-        pass
-
-# PyTorch nn.Module
-if framework == "generic":
-    try:
-        import torch
-        if isinstance(obj, torch.nn.Module):
-            framework = "pytorch"
-    except Exception:
-        pass
-
-# XGBoost
-if framework == "generic":
-    try:
-        import xgboost as xgb
-        if isinstance(obj, (xgb.XGBModel, xgb.Booster)):
-            framework = "xgboost"
-    except Exception:
-        pass
-
-# LightGBM
-if framework == "generic":
-    try:
-        import lightgbm as lgb
-        if isinstance(obj, (lgb.Booster, lgb.LGBMModel)):
-            framework = "lightgbm"
-    except Exception:
-        pass
-
-# CatBoost
-if framework == "generic":
-    try:
-        from catboost import CatBoost
-        if isinstance(obj, CatBoost):
-            framework = "catboost"
-    except Exception:
-        pass
+    import pickle
 
-# sklearn (last — many frameworks inherit from sklearn base)
-if framework == "generic" and "sklearn" in module:
-    framework = "sklearn"
-
-# ── metadata extraction ─────────────────────────────────────────────────────
-class_hierarchy = []
-input_hint = "unknown"
-output_hint = "unknown"
-feature_count = None
-class_labels = None
-
-if framework == "sklearn":
-    if hasattr(obj, "steps"):
-        class_hierarchy = [type(s).__name__ for _, s in obj.steps]
-    else:
-        class_hierarchy = [class_name]
-
-    if hasattr(obj, "n_features_in_"):
-        feature_count = int(obj.n_features_in_)
-
-    if hasattr(obj, "classes_"):
-        class_labels = obj.classes_.tolist()
-    elif hasattr(obj, "steps"):
-        for _, step in obj.steps:
-            if hasattr(step, "classes_"):
-                class_labels = step.classes_.tolist()
-                break
-
-    text_vectorizers = ("TfidfVectorizer", "CountVectorizer", "HashingVectorizer")
-    if class_hierarchy and class_hierarchy[0] in text_vectorizers:
-        input_hint = "raw text string"
-    elif feature_count:
-        input_hint = "array-like of shape (n, " + str(feature_count) + ")"
-    else:
-        input_hint = "array-like"
-
-    if class_labels is not None:
-        output_hint = "integer class label (classes: " + str(class_labels) + ")"
-    else:
-        output_hint = "float or array"
-
-elif framework == "pytorch":
-    try:
-        import torch
-        layer_count = sum(1 for _ in obj.modules())
-        extra["layer_count"] = layer_count
-        # Try to infer input/output shapes from named children
-        children = list(obj.named_children())
-        if children:
-            extra["first_layer"] = type(children[0][1]).__name__
-            extra["last_layer"] = type(children[-1][1]).__name__
-    except Exception:
-        pass
-    input_hint = "torch.Tensor"
-    output_hint = "torch.Tensor"
+    with open(path, "rb") as f:
+        obj = pickle.load(f)
 
-elif framework == "transformers":
-    try:
-        cfg = obj.config
-        extra["model_type"] = getattr(cfg, "model_type", "unknown")
-        extra["hidden_size"] = getattr(cfg, "hidden_size", None)
-        extra["num_labels"] = getattr(cfg, "num_labels", None)
-        extra["tokenizer_class"] = getattr(cfg, "tokenizer_class", None)
-    except Exception:
-        pass
-    input_hint = "dict with input_ids, attention_mask (tokenized)"
-    output_hint = "ModelOutput (logits or last_hidden_state)"
+    module = type(obj).__module__ or ""
+    class_name = type(obj).__name__
 
-elif framework == "xgboost":
-    try:
-        import xgboost as xgb
-        if isinstance(obj, xgb.XGBModel):
-            extra["n_estimators"] = getattr(obj, "n_estimators", None)
-            extra["objective"] = getattr(obj, "objective", None)
-            if hasattr(obj, "n_features_in_"):
-                feature_count = int(obj.n_features_in_)
-        elif isinstance(obj, xgb.Booster):
-            extra["num_trees"] = obj.num_trees()
-    except Exception:
-        pass
-    input_hint = "numpy array or pandas DataFrame"
-    output_hint = "numpy array of predictions"
+    # ── framework detection (order matters) ──────────────────────────────────
+    framework = "generic"
+    extra = {{}}
 
-elif framework == "lightgbm":
+    # sentence-transformers (check before transformers — it wraps HF models)
     try:
-        import lightgbm as lgb
-        if isinstance(obj, lgb.LGBMModel):
-            extra["n_estimators"] = getattr(obj, "n_estimators", None)
-            extra["objective"] = getattr(obj, "objective", None)
-            if hasattr(obj, "n_features_in_"):
-                feature_count = int(obj.n_features_in_)
-        elif isinstance(obj, lgb.Booster):
-            extra["num_trees"] = obj.num_trees()
+        from sentence_transformers import SentenceTransformer
+        if isinstance(obj, SentenceTransformer):
+            framework = "sentence_transformers"
     except Exception:
         pass
-    input_hint = "numpy array or pandas DataFrame"
-    output_hint = "numpy array of predictions"
 
-elif framework == "catboost":
-    try:
-        from catboost import CatBoost
-        extra["loss_function"] = obj.get_param("loss_function")
-        fc = obj.get_param("feature_count") or obj.get_param("num_features")
-        if fc is not None:
-            feature_count = int(fc)
-    except Exception:
-        pass
-    input_hint = "numpy array or pandas DataFrame"
-    output_hint = "numpy array of predictions"
-
-elif framework == "sentence_transformers":
+    # transformers PreTrainedModel
+    if framework == "generic":
+        try:
+            from transformers import PreTrainedModel
+            if isinstance(obj, PreTrainedModel):
+                framework = "transformers"
+        except Exception:
+            pass
+
+    # PyTorch nn.Module
+    if framework == "generic":
+        try:
+            import torch
+            if isinstance(obj, torch.nn.Module):
+                framework = "pytorch"
+        except Exception:
+            pass
+
+    # XGBoost
+    if framework == "generic":
+        try:
+            import xgboost as xgb
+            if isinstance(obj, (xgb.XGBModel, xgb.Booster)):
+                framework = "xgboost"
+        except Exception:
+            pass
+
+    # LightGBM
+    if framework == "generic":
+        try:
+            import lightgbm as lgb
+            if isinstance(obj, (lgb.Booster, lgb.LGBMModel)):
+                framework = "lightgbm"
+        except Exception:
+            pass
+
+    # CatBoost
+    if framework == "generic":
+        try:
+            from catboost import CatBoost
+            if isinstance(obj, CatBoost):
+                framework = "catboost"
+        except Exception:
+            pass
+
+    # sklearn (last — many frameworks inherit from sklearn base)
+    if framework == "generic" and "sklearn" in module:
+        framework = "sklearn"
+
+    # ── metadata extraction ─────────────────────────────────────────────────
+    class_hierarchy = []
+    input_hint = "unknown"
+    output_hint = "unknown"
+    feature_count = None
+    class_labels = None
+
+    if framework == "sklearn":
+        if hasattr(obj, "steps"):
+            class_hierarchy = [type(s).__name__ for _, s in obj.steps]
+        else:
+            class_hierarchy = [class_name]
+
+        if hasattr(obj, "n_features_in_"):
+            feature_count = int(obj.n_features_in_)
+
+        if hasattr(obj, "classes_"):
+            class_labels = obj.classes_.tolist()
+        elif hasattr(obj, "steps"):
+            for _, step in obj.steps:
+                if hasattr(step, "classes_"):
+                    class_labels = step.classes_.tolist()
+                    break
+
+        text_vectorizers = ("TfidfVectorizer", "CountVectorizer", "HashingVectorizer")
+        if class_hierarchy and class_hierarchy[0] in text_vectorizers:
+            input_hint = "raw text string"
+        elif feature_count:
+            input_hint = "array-like of shape (n, " + str(feature_count) + ")"
+        else:
+            input_hint = "array-like"
+
+        if class_labels is not None:
+            output_hint = "integer class label (classes: " + str(class_labels) + ")"
+        else:
+            output_hint = "float or array"
+
+    elif framework == "pytorch":
+        try:
+            import torch
+            layer_count = sum(1 for _ in obj.modules())
+            extra["layer_count"] = layer_count
+            children = list(obj.named_children())
+            if children:
+                extra["first_layer"] = type(children[0][1]).__name__
+                extra["last_layer"] = type(children[-1][1]).__name__
+        except Exception:
+            pass
+        input_hint = "torch.Tensor"
+        output_hint = "torch.Tensor"
+
+    elif framework == "transformers":
+        try:
+            cfg = obj.config
+            extra["model_type"] = getattr(cfg, "model_type", "unknown")
+            extra["hidden_size"] = getattr(cfg, "hidden_size", None)
+            extra["num_labels"] = getattr(cfg, "num_labels", None)
+            extra["tokenizer_class"] = getattr(cfg, "tokenizer_class", None)
+        except Exception:
+            pass
+        input_hint = "dict with input_ids, attention_mask (tokenized)"
+        output_hint = "ModelOutput (logits or last_hidden_state)"
+
+    elif framework == "xgboost":
+        try:
+            import xgboost as xgb
+            if isinstance(obj, xgb.XGBModel):
+                extra["n_estimators"] = getattr(obj, "n_estimators", None)
+                extra["objective"] = getattr(obj, "objective", None)
+                if hasattr(obj, "n_features_in_"):
+                    feature_count = int(obj.n_features_in_)
+            elif isinstance(obj, xgb.Booster):
+                extra["num_trees"] = obj.num_trees()
+        except Exception:
+            pass
+        input_hint = "numpy array or pandas DataFrame"
+        output_hint = "numpy array of predictions"
+
+    elif framework == "lightgbm":
+        try:
+            import lightgbm as lgb
+            if isinstance(obj, lgb.LGBMModel):
+                extra["n_estimators"] = getattr(obj, "n_estimators", None)
+                extra["objective"] = getattr(obj, "objective", None)
+                if hasattr(obj, "n_features_in_"):
+                    feature_count = int(obj.n_features_in_)
+            elif isinstance(obj, lgb.Booster):
+                extra["num_trees"] = obj.num_trees()
+        except Exception:
+            pass
+        input_hint = "numpy array or pandas DataFrame"
+        output_hint = "numpy array of predictions"
+
+    elif framework == "catboost":
+        try:
+            from catboost import CatBoost
+            extra["loss_function"] = obj.get_param("loss_function")
+            fc = obj.get_param("feature_count") or obj.get_param("num_features")
+            if fc is not None:
+                feature_count = int(fc)
+        except Exception:
+            pass
+        input_hint = "numpy array or pandas DataFrame"
+        output_hint = "numpy array of predictions"
+
+    elif framework == "sentence_transformers":
+        try:
+            extra["model_name"] = getattr(obj, "_model_card_text", None) or class_name
+            modules = list(obj.modules())
+            for m in modules:
+                if hasattr(m, "word_embedding_dimension"):
+                    extra["embedding_dim"] = m.word_embedding_dimension
+                    break
+        except Exception:
+            pass
+        input_hint = "string or list of strings"
+        output_hint = "numpy array of shape (n, embedding_dim)"
+
+    artifact_size_mb = round(os.path.getsize(path) / (1024 * 1024), 2)
+    print(json.dumps(dict(
+        framework=framework,
+        class_name=class_name,
+        class_hierarchy=class_hierarchy,
+        input_hint=input_hint,
+        output_hint=output_hint,
+        feature_count=feature_count,
+        class_labels=class_labels,
+        artifact_path=path,
+        artifact_size_mb=artifact_size_mb,
+        extra=extra,
+    )))
+except Exception as _exc:
+    _size = 0.0
     try:
-        extra["model_name"] = getattr(obj, "_model_card_text", None) or class_name
-        modules = list(obj.modules())
-        # embedding dim from the pooling layer
-        for m in modules:
-            if hasattr(m, "word_embedding_dimension"):
-                extra["embedding_dim"] = m.word_embedding_dimension
-                break
+        _size = round(os.path.getsize(path) / (1024 * 1024), 2)
     except Exception:
         pass
-    input_hint = "string or list of strings"
-    output_hint = "numpy array of shape (n, embedding_dim)"
-
-artifact_size_mb = round(os.path.getsize(path) / (1024 * 1024), 2)
-
-result = dict(
-    framework=framework,
-    class_name=class_name,
-    class_hierarchy=class_hierarchy,
-    input_hint=input_hint,
-    output_hint=output_hint,
-    feature_count=feature_count,
-    class_labels=class_labels,
-    artifact_path=path,
-    artifact_size_mb=artifact_size_mb,
-    extra=extra,
-)
-print(json.dumps(result))
+    print(json.dumps(dict(
+        framework="unknown",
+        class_name="unknown",
+        class_hierarchy=[],
+        input_hint="unknown",
+        output_hint="unknown",
+        feature_count=None,
+        class_labels=None,
+        artifact_path=path,
+        artifact_size_mb=_size,
+        extra={{"inspection_warning": str(_exc), "traceback": _tb.format_exc()}},
+    )))
 """
 
 
@@ -284,4 +299,12 @@ def inspect_artifact(path: str) -> ArtifactMetadata:
         raise ValueError(f"Inspection failed:\n{result.stderr.strip()}")
 
     data = json.loads(result.stdout)
-    return ArtifactMetadata(**data)
+    meta = ArtifactMetadata(**data)
+    if "inspection_warning" in meta.extra:
+        import warnings
+        warnings.warn(
+            f"Artifact inspection encountered an error and returned partial metadata: "
+            f"{meta.extra['inspection_warning']}",
+            stacklevel=2,
+        )
+    return meta
diff --git a/docs/internal/issue-planner.md b/docs/internal/issue-planner.md
new file mode 100644
index 0000000..be793a8
--- /dev/null
+++ b/docs/internal/issue-planner.md
@@ -0,0 +1,43 @@
+# Issue Planner
+
+One branch per issue. Branch naming: `fix/<number>-<short-description>` for bugs, `feat/<number>-<short-description>` for enhancements.
+
+---
+
+## Tier 1 — Bugs (fix in order)
+
+| # | Title | Branch | Status |
+|---|-------|--------|--------|
+| #14 | Inspector exits non-zero and discards all metadata on any exception | `fix/14-inspector-exception-handling` | ✅ Done |
+| #15 | Inspector uses pickle.load for all formats, crashing on ONNX and PyTorch `.pt` | `fix/15-inspector-format-routing` | |
+| #21 | `_parse_methods` conflates load() and predict() into one block | `fix/21-parse-methods` | |
+| #20 | `_splice_methods` regex corrupts definition.py when class has helper methods | `fix/20-splice-methods` | |
+| #23 | Reused temp dir across validation retries causes stale module state | `fix/23-validator-stale-module` | |
+| #18 | sample_input passed as raw string to validate_pipeline, breaking numeric models | `fix/18-sample-input-type` | |
+| #22 | write_scaffold crashes with KeyError when ArtifactMetadata fields are None | `fix/22-write-scaffold-none-fields` | |
+| #19 | fix command exits immediately in CI / non-interactive mode | `fix/19-fix-command-ci-mode` | |
+
+## Tier 2 — Enhancements (after all bugs are closed)
+
+| # | Title | Branch | Status |
+|---|-------|--------|--------|
+| #16 | Add LLM interpretation stage between inspection and codegen | `feat/16-llm-interpretation-stage` | |
+| #17 | Add --framework flag to deploy to override framework detection | `feat/17-framework-flag` | |
+| #24 | Add --yes flag for CI mode | `feat/24-yes-flag` | |
+| #25 | Include sample_input in generate(), fix(), and interpretation prompts | `feat/25-sample-input-in-prompts` | |
+| #36 | Extend LLM interpretation to suggest sample_input | `feat/36-suggest-sample-input` | |
+| #35 | Cache LLM generation results by artifact hash | `feat/35-llm-cache` | |
+| #34 | Add GET /models/{name}/{version}/metadata endpoint | `feat/34-metadata-endpoint` | |
+| #33 | Add inference-engine benchmark command | `feat/33-benchmark-command` | |
+| #32 | Add --dry-run flag to fix command | `feat/32-fix-dry-run` | |
+| #31 | Add POST /admin/reload endpoint | `feat/31-admin-reload` | |
+| #30 | Add --explain flag to deploy | `feat/30-explain-flag` | |
+| #28 | Add inference-engine snippets command | `feat/28-snippets-command` | |
+| #27 | Add inference-engine export command | `feat/27-export-command` | |
+| #26 | Add inference-engine package command | `feat/26-package-command` | |
+
+## Tier 3 — Security
+
+| # | Title | Branch | Status |
+|---|-------|--------|--------|
+| #42 | Path traversal in deploy via --name / --version flags | `fix/42-path-traversal` | |
diff --git a/tests/test_cli_phase1_2.py b/tests/test_cli_phase1_2.py
index 17a049b..3a14179 100644
--- a/tests/test_cli_phase1_2.py
+++ b/tests/test_cli_phase1_2.py
@@ -40,11 +40,16 @@ def test_inspect_missing_file():
 
 
 def test_inspect_invalid_file(tmp_path):
+    import warnings
     from app.cli.core.inspector import inspect_artifact
     bad = tmp_path / "bad.pkl"
     bad.write_bytes(b"not a pickle")
-    with pytest.raises(ValueError, match="Inspection failed"):
-        inspect_artifact(str(bad))
+    with warnings.catch_warnings(record=True) as w:
+        warnings.simplefilter("always")
+        meta = inspect_artifact(str(bad))
+    assert meta.framework == "unknown"
+    assert "inspection_warning" in meta.extra
+    assert len(w) == 1 and "partial metadata" in str(w[0].message)
 
 
 # ---------------------------------------------------------------------------