Close audit-writers blind-spot + lock in growth_media id patterns (#85)

Two small follow-ups to the recent schema-gap audit (#84):

- F7: scripts/audit_writers.py's writer-detection heuristic only caught
  yaml.dump / yaml.safe_dump / path.write_text(yaml.dump(...)) patterns.
  scripts/clean_metals_inplace.py mutates 149 community YAMLs via
  regex/line-edit substitutions + path.write_text(string), bypassing
  the detector entirely. Broaden the heuristic: a file that references
  the kb/communities path literal AND round-trips the same path
  variable through .read_text() and .write_text(...) is also a
  community-YAML writer. clean_metals_inplace.py is the only script
  that matches this pattern today; it now surfaces in the audit TSV
  (correctly flagged validates_before_write=no,
  appends_curation_history=no — slated for conversion as separate
  work).

- Fix #5: GrowthMedia.culturemech_id and
  GrowthMediaComponent.media_ingredient_mech_id had no pattern
  declared, while the sibling RelatedMedia.culturemech_id /
  RelatedIngredient.mediaingredientmech_id slots do. Live data
  already conforms to the standard ^CultureMech:\d{6}$ /
  ^MediaIngredientMech:\d{6}$ format — just lock it in to catch
  future drift at write time. Schema regenerated via `just gen-python`;
  `validate_strict.py` still reports 0 ERROR rows across 265 files.

This PR does NOT convert clean_metals_inplace.py itself — that's a
separate, more invasive change (regex surgery → load/dump roundtrip
across 149 files). The audit-blind-spot fix above ensures the next
audit cycle surfaces it.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: realmarcin

reports/pipeline_writers_audit.tsv

@@ -5,6 +5,7 @@ scripts/apply_pmc_conversions.py	yes	yes	yes	yes	no
 scripts/apply_strain_designations.py	yes	no	no	no	no
 scripts/apply_taxonomy_corrections.py	yes	no	no	no	no
 scripts/backfill_metals.py	yes	no	yes	no	no
+scripts/clean_metals_inplace.py	yes	no	yes	no	no
 scripts/enhance_strain_data.py	yes	no	no	no	no
 scripts/fix_network_integrity.py	yes	yes	yes	yes	no
 scripts/fix_reference_formats.py	yes	no	yes	no	no

scripts/audit_writers.py

@@ -38,6 +38,12 @@
 
 # Patterns
 _WRITE_TEXT_OF_YAML = re.compile(r"\.write_text\s*\(\s*yaml\.(?:safe_)?dump")
+# Match the path literal "kb/communities" appearing in source — used to
+# anchor the in-place-mutation heuristic below to scripts that actually
+# operate on community YAMLs.
+_COMMUNITIES_PATH_LITERAL = re.compile(r"kb/communities")
+_READ_TEXT_VAR = re.compile(r"(\w+)\.read_text\s*\(")
+_WRITE_TEXT_VAR = re.compile(r"(\w+)\.write_text\s*\(")
 _CURATION_APPEND = re.compile(
     r"curation_history.*?(append|\+=|\.insert)"
     r"|['\"]curator['\"]\s*:"
@@ -73,7 +79,20 @@ def looks_like_yaml_writer(text: str) -> bool:
         return True
     # write_validated_community is the closed-schema-gated wrapper that
     # callers route through instead of yaml.dump directly.
-    return "write_validated_community(" in text
+    if "write_validated_community(" in text:
+        return True
+    # Text-templated in-place writers: scripts that round-trip the same
+    # path variable through .read_text() and .write_text(...) AND
+    # reference the kb/communities directory are mutating community
+    # YAMLs via raw text surgery (e.g. clean_metals_inplace.py uses
+    # regex/line edits + path.write_text(string)). They bypass yaml.dump
+    # but are still community-YAML writers and must be audited.
+    if _COMMUNITIES_PATH_LITERAL.search(text):
+        read_vars = set(_READ_TEXT_VAR.findall(text))
+        write_vars = set(_WRITE_TEXT_VAR.findall(text))
+        if read_vars & write_vars:
+            return True
+    return False
 
 
 def audit(path: Path, justfile_text: str) -> dict | None:

src/communitymech/datamodel/communitymech.py

@@ -1,5 +1,5 @@
 # Auto generated from communitymech.yaml by pythongen.py version: 0.0.1
-# Generation date: 2026-05-25T01:44:55
+# Generation date: 2026-05-25T18:12:41
 # Schema: communitymech
 #
 # id: https://w3id.org/communitymech
@@ -1949,7 +1949,8 @@ class slots:
                    model_uri=COMMUNITYMECH.growthMediaComponent__name, domain=None, range=str)
 
 slots.growthMediaComponent__media_ingredient_mech_id = Slot(uri=COMMUNITYMECH.media_ingredient_mech_id, name="growthMediaComponent__media_ingredient_mech_id", curie=COMMUNITYMECH.curie('media_ingredient_mech_id'),
-                   model_uri=COMMUNITYMECH.growthMediaComponent__media_ingredient_mech_id, domain=None, range=Optional[str])
+                   model_uri=COMMUNITYMECH.growthMediaComponent__media_ingredient_mech_id, domain=None, range=Optional[str],
+                   pattern=re.compile(r'^MediaIngredientMech:\d{6}$'))
 
 slots.growthMediaComponent__media_ingredient_mech_url = Slot(uri=COMMUNITYMECH.media_ingredient_mech_url, name="growthMediaComponent__media_ingredient_mech_url", curie=COMMUNITYMECH.curie('media_ingredient_mech_url'),
                    model_uri=COMMUNITYMECH.growthMediaComponent__media_ingredient_mech_url, domain=None, range=Optional[str])
@@ -1970,7 +1971,8 @@ class slots:
                    model_uri=COMMUNITYMECH.growthMedia__name, domain=None, range=str)
 
 slots.growthMedia__culturemech_id = Slot(uri=COMMUNITYMECH.culturemech_id, name="growthMedia__culturemech_id", curie=COMMUNITYMECH.curie('culturemech_id'),
-                   model_uri=COMMUNITYMECH.growthMedia__culturemech_id, domain=None, range=Optional[str])
+                   model_uri=COMMUNITYMECH.growthMedia__culturemech_id, domain=None, range=Optional[str],
+                   pattern=re.compile(r'^CultureMech:\d{6}$'))
 
 slots.growthMedia__culturemech_url = Slot(uri=COMMUNITYMECH.culturemech_url, name="growthMedia__culturemech_url", curie=COMMUNITYMECH.curie('culturemech_url'),
                    model_uri=COMMUNITYMECH.growthMedia__culturemech_url, domain=None, range=Optional[str])

src/communitymech/schema/communitymech.yaml

@@ -699,6 +699,7 @@ classes:
         required: true
       media_ingredient_mech_id:
         description: Identifier in MediaIngredientMech database (e.g., MediaIngredientMech:000001)
+        pattern: "^MediaIngredientMech:\\d{6}$"
       media_ingredient_mech_url:
         description: URL to MediaIngredientMech ingredient entry
       concentration:
@@ -719,6 +720,7 @@ classes:
         required: true
       culturemech_id:
         description: Identifier in CultureMech media database (e.g., CultureMech:000001)
+        pattern: "^CultureMech:\\d{6}$"
       culturemech_url:
         description: URL to CultureMech media entry
       composition: