Adding file path + exploit scenario part

Author: kevin-valerio

README.md

@@ -9,6 +9,8 @@ Different is a variant-analysis agentic tool built with DeepAgents. It does two
 
 The logic is agentic: an LLM calls local Git tools (and optional GitHub API tools) in a loop to inspect commits, diffs, and related PR/issue context.
 
+Each finding includes `id`, `kind`, `severity`, `title`, `root_cause`, `fix_summary`, `evidence`, and `tags`. For `kind="bug"` findings with a concrete severity (not `"unknown"`), it also includes `main_file` and `exploit_risk` (a short paragraph describing how an attacker could exploit the bug and what impact they could get).
+
 The inspiration agent can fetch PR labels and review comments for richer context. The target agent can search commit messages (`git_log_search`) to check if a fix was already applied, and list tracked files (`git_ls_files`) to explore the project structure.
 
 ## When to use it

src/different_agent/agents.py

@@ -6,7 +6,7 @@
 from langchain.agents.structured_output import AutoStrategy
 from langchain_core.language_models import BaseChatModel
 from langgraph.cache.base import BaseCache
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 
 from different_agent.git_tools import (
     ast_grep,
@@ -28,7 +28,7 @@
     github_recent_prs,
 )
 
-FINDING_SCHEMA_VERSION = "v1"
+FINDING_SCHEMA_VERSION = "v3"
 
 
 class EvidenceCommit(BaseModel):
@@ -49,11 +49,25 @@ class Finding(BaseModel):
     kind: Literal["bug", "vulnerability", "hardening"]
     title: str
     severity: Literal["low", "medium", "high", "critical", "unknown"]
+    main_file: str | None = None
+    exploit_risk: str | None = None
     root_cause: str
     fix_summary: str
     evidence: FindingEvidence
     tags: list[str] = Field(default_factory=list)
 
+    @model_validator(mode="after")
+    def _require_bug_risk_fields(self) -> Finding:
+        if self.kind != "bug" or self.severity == "unknown":
+            return self
+        if not self.main_file:
+            raise ValueError("main_file is required for kind='bug' with severity!='unknown'")
+        if not self.exploit_risk or not self.exploit_risk.strip():
+            raise ValueError("exploit_risk is required for kind='bug' with severity!='unknown'")
+        if self.exploit_risk.strip().lower() in {"low", "medium", "high", "critical", "unknown"}:
+            raise ValueError("exploit_risk must be a short paragraph, not a keyword")
+        return self
+
 
 class FindingsResponse(BaseModel):
     findings: list[Finding]
@@ -105,6 +119,9 @@ class TargetAssessmentsResponse(BaseModel):
 - Do NOT paste entire diffs into the JSON. Keep diff_snippets short.
 - If you include GitHub issues/PRs, include their links in evidence.links.
 - Be conservative: if you can't justify severity, set severity="unknown".
+- If kind="bug" and severity!="unknown":
+  - main_file must be set to the single most relevant file path for the fix (usually pick 1 entry from evidence.files_changed).
+  - exploit_risk must be a short paragraph explaining how an attacker could exploit the bug and what they could gain (preconditions + impact). Keep it similar length to root_cause/fix_summary.
 - root_cause must describe the generalized mechanism (unsafe pattern + conditions), not just the local symbol name.
 - fix_summary must describe the conceptual mitigation, not only the exact code change.
 - tags should include short idea-level keywords to help variant matching (e.g. "ambiguous-encoding",
@@ -120,6 +137,8 @@ class TargetAssessmentsResponse(BaseModel):
 - kind ("bug" | "vulnerability" | "hardening")
 - title (string)
 - severity ("low" | "medium" | "high" | "critical" | "unknown")
+- main_file (string|null)  # required when kind="bug" and severity!="unknown"
+- exploit_risk (string|null)  # required when kind="bug" and severity!="unknown"
 - root_cause (string)
 - fix_summary (string)
 - evidence {{

src/different_agent/report.py

@@ -17,6 +17,11 @@ def render_findings_html(findings: list[dict]) -> str:
     now = datetime.now(UTC).replace(microsecond=0).isoformat()
     rows = []
     for f in findings:
+        kind = f.get("kind")
+        severity = f.get("severity")
+        show_risk_fields = kind == "bug" and severity not in (None, "", "unknown")
+        main_file = f.get("main_file") if show_risk_fields else ""
+        exploit_risk = f.get("exploit_risk") if show_risk_fields else ""
         rows.append(
             "<tr>"
             f"<td>{_safe_json(f.get('id'))}</td>"
@@ -25,6 +30,8 @@ def render_findings_html(findings: list[dict]) -> str:
             f"<td>{_safe_json(f.get('title'))}</td>"
             f"<td><pre>{_safe_json(f.get('root_cause'))}</pre></td>"
             f"<td><pre>{_safe_json(f.get('fix_summary'))}</pre></td>"
+            f"<td>{_safe_json(main_file)}</td>"
+            f"<td><pre>{_safe_json(exploit_risk)}</pre></td>"
             "</tr>"
         )
 
@@ -58,6 +65,8 @@ def render_findings_html(findings: list[dict]) -> str:
           <th>title</th>
           <th>root_cause</th>
           <th>fix_summary</th>
+          <th>main_file</th>
+          <th>exploit_risk</th>
         </tr>
       </thead>
       <tbody>

tests/test_report.py

@@ -10,6 +10,8 @@ def test_render_findings_html_escapes_values() -> None:
             "kind": "bug",
             "severity": "high",
             "title": "Title & stuff",
+            "main_file": "src/<main>.py",
+            "exploit_risk": "<high>",
             "root_cause": "a < b",
             "fix_summary": "use & sanitize",
         }
@@ -18,6 +20,8 @@ def test_render_findings_html_escapes_values() -> None:
     assert "Findings" in html
     assert "&lt;f-1&gt;" in html
     assert "Title &amp; stuff" in html
+    assert "src/&lt;main&gt;.py" in html
+    assert "&lt;high&gt;" in html
     assert "a &lt; b" in html
     assert "use &amp; sanitize" in html