v0.1.29 feat: intent capture via set_intent MCP tool

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "agentdiff"
-version = "0.1.28"
+version = "0.1.29"
 edition = "2024"
 rust-version = "1.85"
 description = "Audit and trace autonomous AI code contributions in git repositories"

scripts/finalize-ledger.py

@@ -63,6 +63,43 @@ def remove_if_exists(path: str) -> None:
         pass
 
 
+def remove_consumed_intents(session_path: str) -> None:
+    """Strip type=intent events from session.jsonl after they've been committed.
+
+    Intent events are one-shot: set_intent is called just before a commit, so
+    by the time finalize runs they've been consumed. Leaving them causes stale
+    intent to bleed into the next commit when two commits share the same second.
+    """
+    if not os.path.exists(session_path):
+        return
+    kept = []
+    try:
+        with open(session_path, "r", encoding="utf-8") as f:
+            for raw in f:
+                line = raw.strip()
+                if not line:
+                    continue
+                try:
+                    event = json.loads(line)
+                    if isinstance(event, dict) and event.get("type") == "intent":
+                        continue
+                except Exception:
+                    pass
+                kept.append(raw if raw.endswith("\n") else raw + "\n")
+    except (OSError, IOError):
+        return
+    tmp = session_path + ".tmp"
+    try:
+        with open(tmp, "w", encoding="utf-8") as f:
+            f.writelines(kept)
+        os.replace(tmp, session_path)
+    except (OSError, IOError):
+        try:
+            os.remove(tmp)
+        except OSError:
+            pass
+
+
 def sha_already_recorded(traces_path: str, sha: str) -> bool:
     """Skip finalize if this commit already has a trace recorded locally."""
     if not os.path.exists(traces_path):
@@ -158,6 +195,8 @@ def write_agent_trace(repo_root: str, pending: dict, sha: str, ts: str) -> Optio
         metadata["session_id"] = str(pending["session_id"])
     if pending.get("intent"):
         metadata["intent"] = str(pending["intent"])
+    if pending.get("intent_type"):
+        metadata["intent_type"] = str(pending["intent_type"])
     if isinstance(pending.get("files_read"), list) and pending["files_read"]:
         metadata["files_read"] = [str(p) for p in pending["files_read"]]
     if git_author:
@@ -228,9 +267,12 @@ def main() -> int:
         return 1
     ts = ts_res.stdout.strip()
 
+    session_path = os.path.join(repo_root, ".git", "agentdiff", "session.jsonl")
     result = write_agent_trace(repo_root, pending, sha, ts)
     remove_if_exists(pending_ledger_path)
     remove_if_exists(pending_context_path)
+    if result is not None:
+        remove_consumed_intents(session_path)
     return 0 if result is not None else 1
 
 

scripts/prepare-ledger.py

@@ -198,6 +198,41 @@ def read_events_per_file(
     return {fp: ev for fp, (_, ev) in best.items()}
 
 
+def read_intent_events(
+    path: str,
+    min_ts: int,
+) -> List[dict]:
+    """Read intent events from session.jsonl (type=intent, written by set_intent MCP tool).
+
+    Returns all matching intent events sorted by timestamp (most recent last).
+    """
+    if not os.path.exists(path):
+        return []
+    results = []
+    try:
+        with open(path, "r", encoding="utf-8") as f:
+            for raw in f:
+                line = raw.strip()
+                if not line:
+                    continue
+                try:
+                    event = json.loads(line)
+                except Exception:
+                    continue
+                if not isinstance(event, dict):
+                    continue
+                if event.get("type") != "intent":
+                    continue
+                event_ts = parse_event_ts(str(event.get("timestamp") or ""))
+                if event_ts < min_ts:
+                    continue
+                results.append(event)
+    except Exception:
+        pass
+    results.sort(key=lambda e: parse_event_ts(str(e.get("timestamp") or "")))
+    return results
+
+
 def dominant_event(events_by_file: Dict[str, dict], lines_by_file: Dict[str, List[Tuple[int, int]]]) -> dict:
     """Pick the agent/model to use as the top-level record field.
 
@@ -276,13 +311,17 @@ def main() -> int:
     if not isinstance(pending, dict):
         pending = {}
 
+    min_ts = head_commit_ts(repo_root)
     events_by_file = read_events_per_file(
         session_log,
         files_touched,
-        head_commit_ts(repo_root),
+        min_ts,
         lines_by_file,
     )
 
+    # Read agent-stated intent events from session.jsonl (written by set_intent MCP tool)
+    intent_events = read_intent_events(session_log, min_ts)
+
     # Top-level agent/model/session come from the dominant event (most lines written)
     event = dominant_event(events_by_file, lines_by_file) or {}
 
@@ -307,9 +346,31 @@ def main() -> int:
         flags = []
     flags = [str(f) for f in flags]
 
-    intent = pending.get("intent")
-    if intent is not None:
-        intent = str(intent)
+    # Intent priority: agent-stated intent event > pending context > fallback
+    intent = None
+    intent_type = None
+    if intent_events:
+        # Use the most recent intent event (prefer matching session, else latest)
+        best_intent = None
+        if session_id and session_id != "unknown":
+            for ie in reversed(intent_events):
+                if str(ie.get("session_id") or "") == session_id:
+                    best_intent = ie
+                    break
+        if not best_intent:
+            best_intent = intent_events[-1]
+        intent = str(best_intent.get("description") or "").strip() or None
+        intent_type = str(best_intent.get("intent_type") or "").strip() or None
+
+    if not intent:
+        raw = pending.get("intent")
+        if raw is not None:
+            intent = str(raw).strip() or None
+
+    if not intent_type:
+        raw = pending.get("intent_type")
+        if raw is not None:
+            intent_type = str(raw).strip() or None
 
     # Per-file attribution — each file maps to the agent/model that wrote it.
     # Files with a session event that matches the dominant agent are omitted (finalize
@@ -364,6 +425,8 @@ def main() -> int:
         payload["attribution"] = attribution
     if intent:
         payload["intent"] = intent
+    if intent_type:
+        payload["intent_type"] = intent_type
     if trust is not None:
         payload["trust"] = trust
 

scripts/record-context.py

@@ -64,6 +64,9 @@ def main() -> int:
     parser.add_argument("--prompt", default="")
     parser.add_argument("--files-read", default="")
     parser.add_argument("--intent", default="")
+    parser.add_argument("--intent-type", default="",
+                        choices=["bugfix","feature","refactor","test","docs",
+                                 "security","performance","config","dependency",""])
     parser.add_argument("--trust", type=int, default=None)
     parser.add_argument("--flags", default="")
     args = parser.parse_args()
@@ -90,6 +93,7 @@ def main() -> int:
         "prompt": args.prompt or str(payload.get("prompt") or ""),
         "files_read": parse_json_array(args.files_read) or payload.get("files_read") or [],
         "intent": args.intent or str(payload.get("intent") or ""),
+        "intent_type": args.intent_type or str(payload.get("intent_type") or ""),
         "flags": parse_json_array(args.flags) or payload.get("flags") or [],
     }
 

scripts/tests/test_capture_prompts.py

@@ -208,6 +208,53 @@ def test_write_agent_trace_persists_structured_context_metadata(self):
             self.assertEqual(metadata["author"], "Prakhar")
             self.assertEqual(metadata["capture_tool"], "afterFileEdit")
 
+    def test_write_agent_trace_persists_intent_type(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            repo = Path(tmp) / "repo"
+            repo.mkdir()
+            subprocess.run(["git", "init", "-b", "main"], cwd=repo, check=True, capture_output=True)
+            subprocess.run(["git", "config", "user.email", "test@example.com"], cwd=repo, check=True)
+            subprocess.run(["git", "config", "user.name", "Test User"], cwd=repo, check=True)
+            (repo / "README.md").write_text("test\n", encoding="utf-8")
+            subprocess.run(["git", "add", "README.md"], cwd=repo, check=True)
+            subprocess.run(["git", "commit", "-m", "init"], cwd=repo, check=True, capture_output=True)
+
+            pending = {
+                "agent": "cursor",
+                "git_author": "Prakhar",
+                "model": "cursor-test",
+                "session_id": "sess-2",
+                "lines": {"src/app.py": [[1, 5]]},
+                "prompt_excerpt": "extract auth middleware",
+                "prompt_hash": "def456",
+                "intent": "eliminate duplicate token validation across route handlers",
+                "intent_type": "refactor",
+                "files_read": [],
+                "trust": 85,
+                "flags": [],
+                "tool": "afterFileEdit",
+            }
+
+            original = os.environ.get("HOME")
+            try:
+                os.environ["HOME"] = tmp
+                traces_path = self.mod.write_agent_trace(
+                    str(repo), pending, "cafebabe", "2026-05-21T00:00:00Z"
+                )
+            finally:
+                if original is not None:
+                    os.environ["HOME"] = original
+
+            self.assertIsNotNone(traces_path)
+            raw = Path(traces_path).read_text(encoding="utf-8").strip()
+            trace = json.loads(raw)
+            metadata = trace["metadata"]["agentdiff"]
+            self.assertEqual(metadata["intent_type"], "refactor")
+            self.assertEqual(
+                metadata["intent"],
+                "eliminate duplicate token validation across route handlers",
+            )
+
 
 if __name__ == "__main__":
     unittest.main()