Tag fenced blocks with language for markdownlint, widen parser fence regex

bin/plugin-inventory

@@ -46,7 +46,7 @@ def load_catalog(path):
         name = m.group(1)
         if name not in KNOWN_SECTIONS:
             continue
-        block = re.search(r"```\n(.*?)```", text[m.end():], re.DOTALL)
+        block = re.search(r"```\S*\n(.*?)```", text[m.end():], re.DOTALL)
         if not block:
             continue
         body = block.group(1)

references/signals-catalog.md

@@ -19,7 +19,7 @@ literal `#` at the start of a line.
 Occurrences count toward the script-verb signal. Matched case-insensitively,
 whole-word, against skill/agent body prose (frontmatter excluded).
 
-```
+```text
 count, sum, rank, sort, diff, parse, validate, enumerate, dedupe, dedup,
 deduplicate, format, compare, aggregate, glob, grep, filter, normalize,
 normalise, tabulate, checksum, hash, lint, tally, group, join, merge, split,
@@ -31,7 +31,7 @@ decode
 
 Occurrences count toward the judgment-verb signal. Same matching rule.
 
-```
+```text
 decide, judge, explain, prioritize, prioritise, name, write, weigh,
 interpret, recommend, assess, evaluate, justify, persuade, negotiate,
 empathize, empathise, phrase, rephrase, summarize, summarise, synthesize,
@@ -50,7 +50,7 @@ the extraction is exhaustive.
 Tools whose presence in an agent's `tools:` list leans that agent toward
 "built to fetch." Used for `mechanical_tool_share`.
 
-```
+```text
 Bash, Read, Grep, Glob, WebFetch, WebSearch, LS, NotebookRead,
 ListMcpResourcesTool, ReadMcpResourceTool, ReadMcpResourceDirTool, TaskGet,
 TaskList, TaskOutput, Monitor
@@ -63,7 +63,7 @@ consequential, T5/P4), AskUserQuestion (is itself an escalation), Agent
 
 ## thresholds
 
-```
+```text
 verb_ratio_high: 0.6
 # verb_ratio (script / (script+judgment)) above this with zero bin/ script
 # coverage is the loudest static smell per spec.md §5.1.