Fix observe parsing for holistic issue IDs

desloppify/app/commands/plan/triage/runner/stage_prompts_observe.py

@@ -87,7 +87,8 @@ def build_observe_batch_prompt(
         title = f.get("title", fid)
         file_path = detail.get("file_path", "")
         description = detail.get("description", f.get("description", ""))
-        line = f"- [{fid[:8]}] ({dim}) **{title}**"
+        issue_token = str(detail.get("summary_hash") or fid.rsplit("::", 1)[-1])
+        line = f"- [{issue_token}] ({dim}) **{title}**"
         if file_path:
             line += f" — `{file_path}`"
         if description:

desloppify/app/commands/plan/triage/stages/evidence_parsing.py

@@ -55,7 +55,7 @@ class ObserveEvidence:
 
     entries: list[ObserveAssessment] = field(default_factory=list)
     unparsed_citation_count: int = 0  # hashes cited without a verdict
-    has_parseable_ids: bool = True  # False if valid_ids had no hex-hash IDs
+    has_parseable_ids: bool = True  # False if valid_ids had no parseable short IDs
 
 
 @dataclass
@@ -93,12 +93,18 @@ def _normalise_verdict(raw: str) -> str | None:
 
 
 def _build_short_map(valid_ids: set[str]) -> dict[str, str]:
-    """Build short-hash → full-id map from valid issue IDs."""
+    """Build collision-safe short-id → full-id map from valid issue IDs."""
     short_map: dict[str, str] = {}
     for fid in valid_ids:
-        short = fid.rsplit("::", 1)[-1]
-        if re.fullmatch(r"[0-9a-f]{6,}", short):
+        short = fid.rsplit("::", 1)[-1].lower()
+        if not short:
+            continue
+        existing = short_map.get(short)
+        if existing is None:
             short_map[short] = fid
+            continue
+        if existing != fid:
+            short_map.pop(short, None)
     return short_map
 
 
@@ -109,8 +115,8 @@ def _is_valid_hash(raw_hash: str, short_map: dict[str, str], valid_ids: set[str]
 
 # --- YAML-like template parser ---
 
-# Matches lines like:  - hash: abc12345  or  hash: abc12345
-_YAML_HASH_RE = re.compile(r"^\s*-?\s*hash\s*:\s*([0-9a-f]{6,})", re.IGNORECASE)
+# Matches lines like:  - hash: abc12345  or  hash: task_param_bag
+_YAML_HASH_RE = re.compile(r"^\s*-?\s*hash\s*:\s*(\S+)", re.IGNORECASE)
 # Matches lines like:  verdict: genuine  or  verdict: false-positive
 _YAML_VERDICT_RE = re.compile(r"^\s*verdict\s*:\s*(.+)", re.IGNORECASE)
 # Matches lines like:  verdict_reasoning: some reason
@@ -214,9 +220,13 @@ def parse_observe_evidence(report: str, valid_ids: set[str]) -> ObserveEvidence:
     short_map = _build_short_map(valid_ids)
     entries = _parse_yaml_template(report, short_map, valid_ids)
 
-    # Count hashes that appear in report but weren't parsed as entries
-    all_hashes_in_report = set(re.findall(r"[0-9a-f]{8,}", report.lower()))
+    # Count short IDs that appear in report but weren't parsed as entries.
     valid_short_hashes = set(short_map.keys())
+    all_hashes_in_report = {
+        match.group(1).lower()
+        for line in report.splitlines()
+        if (match := _YAML_HASH_RE.match(line))
+    }
     cited_hashes = all_hashes_in_report & valid_short_hashes
     parsed_hashes = {e.issue_hash for e in entries}
     unparsed = len(cited_hashes - parsed_hashes)
@@ -250,7 +260,7 @@ def validate_observe_evidence(
     if issue_count == 0:
         return []
 
-    # If valid_ids contained no hex-hash IDs, verdict parsing is impossible — skip.
+    # If valid_ids contained no parseable short IDs, verdict parsing is impossible — skip.
     if not evidence.has_parseable_ids:
         return []
 
@@ -590,6 +600,10 @@ def resolve_short_hash_to_full_id(short_hash: str, valid_ids: set[str]) -> str |
     # Try direct match first
     if short_hash in valid_ids:
         return short_hash
+    short_hash = short_hash.lower()
+    bucket = maps.short_id_buckets.get(short_hash)
+    if bucket and len(bucket) == 1:
+        return bucket[0]
     # Try collision-aware short hex map
     resolved = maps.short_hex_map.get(short_hash)
     if resolved:

desloppify/tests/commands/plan/test_triage_evidence_parsing.py

@@ -144,6 +144,22 @@ def test_yaml_entries_parse_when_legacy_lines_are_present(self):
         assert len(evidence.entries) == 1
         assert evidence.entries[0].issue_hash == "abc12345"
 
+    def test_yaml_template_parses_semantic_holistic_short_ids(self):
+        valid_ids = {"review::.::holistic::naming_quality::task_param_bag"}
+        report = (
+            "- hash: task_param_bag\n"
+            "  verdict: genuine\n"
+            "  verdict_reasoning: Real issue confirmed in src/types.ts.\n"
+            "  files_read: [src/types.ts]\n"
+            "  recommendation: Refactor the shared parameter bag\n"
+        )
+
+        evidence = parse_observe_evidence(report, valid_ids)
+
+        assert evidence.has_parseable_ids is True
+        assert len(evidence.entries) == 1
+        assert evidence.entries[0].issue_hash == "task_param_bag"
+
 
 # ---------------------------------------------------------------------------
 # validate_observe_evidence — field presence checks

desloppify/tests/commands/plan/test_triage_stage_prompts_flow_direct.py

@@ -64,7 +64,15 @@ def test_observe_and_sense_prompt_builders_include_expected_context(tmp_path) ->
                 "title": "Naming issue",
                 "description": "rename to clear name",
                 "detail": {"dimension": "naming_quality", "file_path": "src/a.py"},
-            }
+            },
+            "review::.::holistic::naming_quality::task_param_bag": {
+                "title": "Holistic naming issue",
+                "description": "shared task parameter bag hides intent",
+                "detail": {
+                    "dimension": "naming_quality",
+                    "summary_hash": "1a2b3c4d",
+                },
+            },
         },
         repo_root=tmp_path,
     )
@@ -73,7 +81,8 @@ def test_observe_and_sense_prompt_builders_include_expected_context(tmp_path) ->
     assert "observe batch 1/2" in observe
     assert "naming_quality" in observe
     assert f"Repo root: {tmp_path}" in observe
-    assert "[review::s" not in observe  # hash prefix truncation is used
+    assert "[abcdef12]" in observe
+    assert "[1a2b3c4d]" in observe
     for required_field in ("- hash:", "verdict:", "verdict_reasoning:", "files_read:", "recommendation:"):
         assert required_field in observe
     assert "Do NOT run any `desloppify` commands" in observe