fix(secrets): the scanner printed the credential it found

The report truncated each line to 120 characters and redacted nothing, so a
credential shorter than the budget was printed whole -- into the terminal
scrollback, the CI log, and wherever that log is shipped. It applied to every
mode, which meant a real leak caught by the pre-commit hook was also a real leak
printed to a terminal. The comment above it claimed the match was never echoed
in full; it was corrected to describe the behaviour in the previous commit, and
this changes the behaviour instead.

The match is now masked before truncation. \001 is the substitution delimiter,
as a real control byte rather than the literal backslash-zero-zero-one a
double-quoted "\001" produces -- that first attempt made sed take `\` as its
delimiter and silently substitute nothing, which looked exactly like working
code. These patterns contain both / and |, so either would end the expression
early.

Widening the JWT pattern was part of the same fix, not a separate improvement.
Masking removes exactly what the pattern matched, so `eyJ[A-Za-z0-9_-]{10,}`
redacted the header and printed the payload and signature next to it -- and
those are the token. It now matches all three segments. A pattern that
under-matches is a pattern that half-prints the secret.

Verified in --built and --staged: a planted JWT and a user:pass@host URL are
each reported with file and line, and neither planted value appears anywhere in
the output.

closes #11

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

docs/architecture/scripts/secrets.sh

@@ -68,6 +68,11 @@ cd "$(git rev-parse --show-toplevel 2>/dev/null)" || {
 
 say() { printf 'secrets: %s\n' "$*" >&2; }
 
+# The delimiter for the masking substitution below. A real control byte, because
+# these patterns contain both `/` and `|` and either would end the expression
+# early. It cannot occur in a pattern and it cannot occur in source text.
+MASK_D=$'\001'
+
 MODE="staged"
 BUILT_DIR=""
 ALLOW=()
@@ -92,7 +97,10 @@ done
 # and it is far too eager to run against source, where it matches ordinary
 # base64. In a bundle it is worth the noise.
 BUILT_PATTERNS=(
-  'eyJ[A-Za-z0-9_-]{10,}'
+  # All three segments, not just the header. Masking removes exactly what the
+  # pattern matched, so a pattern that stops at the first dot redacts `eyJ...`
+  # and prints the payload and signature beside it -- which is the token.
+  'eyJ[A-Za-z0-9_-]{10,}(\.[A-Za-z0-9_-]+){0,2}'
   '\bservice_role\b'
   '\bapikey["'"'"'[:space:]]*[:=]'
 )
@@ -225,13 +233,18 @@ for pattern in "${PATTERNS[@]}"; do
     done
     [ -n "$skip" ] && continue
 
-    # The line is truncated to 120 characters, which bounds how much of a long
+    # The match is masked, then the line is truncated. Truncation alone was not
-    # value reaches the terminal. It does NOT redact the match: a credential
+    # enough and used to be all there was: it bounds how much of a LONG value
-    # shorter than the budget is printed whole, and a scanner that prints the
+    # reaches the terminal and prints a short one whole, so the scanner
-    # secret it found has published it to the scrollback, the CI log, and
+    # published the very thing it was built to find — to the scrollback, the CI
-    # wherever that log is shipped. Masking the matched span is the real fix and
+    # log, and wherever that log is shipped.
-    # is tracked separately -- until then, treat this output as sensitive.
+    #
-    printf '  %.120s…\n' "$hit"
+    # \001 as the delimiter, because these patterns contain both `/` and `|`
+    # and either would end the expression early. It cannot occur in a pattern
+    # and it cannot occur in the text of a source file.
+    masked="$(printf '%s' "$hit" | sed -E "s${MASK_D}${pattern}${MASK_D}[redacted]${MASK_D}g" 2>/dev/null)"
+    [ -n "$masked" ] || masked="[a line matching a credential pattern, unprintable]"
+    printf '  %.120s…\n' "$masked"
     found=$((found + 1))
   done < <(printf '%s\n' "$CONTENT" | grep -nEI "$pattern" 2>/dev/null | head -20)
 done