docs/raw-inbox.md

@@ -33,7 +33,7 @@ deployment:
 |---|---|
 | `INGEST_WING_ROUTING` | `1`/`true` enables the rules above; unset or `0`/`false` is off; any other value stops the server at startup |
 | `INGEST_DEFAULT_WING` | wing for unsignalled notes (default `web-clip`; must be a plain name, not `sources`/`concepts`/`entities`/`wiki`) |
-| `INGEST_CLIENT_BLOCK` | case-insensitive regexp over note text; a match parks an unsignalled note. Falls back to `CLAUDE_INGEST_CLIENT_BLOCK`. Matched after folding fullwidth letters, removing zero-width characters, normalising dashes and removing whitespace. Patterns with spaces, `\\s` or `\\b` work (whitespace is collapsed in one view and removed in another). It catches what it names and nothing else: leetspeak, punctuation splits (`a.c.m.e`), encodings and homoglyphs outside a small Cyrillic/Greek table are not caught. Large notes are checked in 1 MiB chunks |
+| `INGEST_CLIENT_BLOCK` | case-insensitive regexp over note text; a match parks an unsignalled note. Falls back to `CLAUDE_INGEST_CLIENT_BLOCK`, which the deployment injects as a bare list of client names (`acme|globex`): it is then matched as whole names, with word edges decided by Unicode letters and digits (an underscore or punctuation is a boundary, so `trucker_ecommerce` matches `trucker`; `acmecorp`, `acme2` and plurals do not), and a list that is not a valid regexp, has an empty alternative, or could break out of its group stops the server at startup. A full pattern that matches the empty string is refused at startup too. Matched after folding fullwidth letters, removing zero-width characters, normalising dashes and removing whitespace. Patterns with spaces, `\\s` or `\\b` work (whitespace is collapsed in one view and removed in another). It catches what it names and nothing else: leetspeak, punctuation splits (`a.c.m.e`), encodings and homoglyphs outside a small Cyrillic/Greek table are not caught. Large notes are checked in 1 MiB chunks |
 
 A bad default or pattern stops the server at startup. A wing created while the
 server runs is usable without a restart.

ingestion/cmd/server/main.go

@@ -150,6 +150,22 @@ func envInt(key string, fallback int) int {
 	return fallback
 }
 
+// ingestClientPattern picks the client-content pattern for raw/ routing.
+// INGEST_CLIENT_BLOCK is a complete regexp and is used as is. Otherwise the
+// shared CLAUDE_INGEST_CLIENT_BLOCK is used: the deployment injects it as a bare
+// list of client names (regexp alternation), which wing.WrapNames bounds so a
+// short name does not match inside ordinary words, and which it refuses if it
+// would break out of its group or match the empty string.
+func ingestClientPattern(getenv func(string) string) (string, error) {
+	if p := getenv("INGEST_CLIENT_BLOCK"); p != "" {
+		return p, nil
+	}
+	if names := getenv("CLAUDE_INGEST_CLIENT_BLOCK"); names != "" {
+		return wing.WrapNames(names)
+	}
+	return "", nil
+}
+
 // envFlag reads a boolean switch. Unset is off; a value strconv.ParseBool does
 // not understand is an error, so a typo ("ture") cannot silently leave a
 // safety-relevant switch off.
@@ -426,9 +442,10 @@ func main() {
 			os.Exit(1)
 		}
 		if routing {
-			pattern := os.Getenv("INGEST_CLIENT_BLOCK")
-			if pattern == "" {
-				pattern = os.Getenv("CLAUDE_INGEST_CLIENT_BLOCK")
+			pattern, perr := ingestClientPattern(os.Getenv)
+			if perr != nil {
+				logger.Error("client-content pattern", "err", perr)
+				os.Exit(1)
 			}
 			router, rerr := wing.NewRouter(brainDir, envOr("INGEST_DEFAULT_WING", "web-clip"), pattern)
 			if rerr != nil {

ingestion/cmd/server/main_test.go

@@ -129,3 +129,30 @@ func TestEnvFlag(t *testing.T) {
 		})
 	}
 }
+
+// The deployment injects CLAUDE_INGEST_CLIENT_BLOCK as a bare list of client
+// names (regexp alternation). Used raw, a short name would match inside ordinary
+// words; wrapped with \b, a name with a non-ASCII edge letter would never match.
+func TestIngestClientPattern(t *testing.T) {
+	get := func(m map[string]string) func(string) string { return func(k string) string { return m[k] } }
+	t.Run("nothing set", func(t *testing.T) {
+		p, err := ingestClientPattern(get(map[string]string{}))
+		assert.NoError(t, err)
+		assert.Empty(t, p)
+	})
+	t.Run("explicit pattern is a full regexp, used as is", func(t *testing.T) {
+		p, err := ingestClientPattern(get(map[string]string{"INGEST_CLIENT_BLOCK": `acme\s+corp`, "CLAUDE_INGEST_CLIENT_BLOCK": "other"}))
+		assert.NoError(t, err)
+		assert.Equal(t, `acme\s+corp`, p)
+	})
+	t.Run("the shared names are wrapped", func(t *testing.T) {
+		p, err := ingestClientPattern(get(map[string]string{"CLAUDE_INGEST_CLIENT_BLOCK": "dif|trucker"}))
+		assert.NoError(t, err)
+		assert.Contains(t, p, "dif|trucker")
+		assert.NotEqual(t, "dif|trucker", p)
+	})
+	t.Run("a shared value that would break out of the group is a startup error", func(t *testing.T) {
+		_, err := ingestClientPattern(get(map[string]string{"CLAUDE_INGEST_CLIENT_BLOCK": "x)|.*("}))
+		assert.Error(t, err)
+	})
+}

ingestion/internal/wing/router.go

@@ -94,6 +94,9 @@ func NewRouter(brainDir, defaultWing, clientPattern string) (Router, error) {
 		if err != nil {
 			return Router{}, fmt.Errorf("client-content pattern: %w", err)
 		}
+		if re.MatchString("x") || re.MatchString("") {
+			return Router{}, fmt.Errorf("client-content pattern %q matches text that names no client; it would park every note", clientPattern)
+		}
 		r.ClientBlock = re
 	}
 	return r, nil
@@ -218,3 +221,28 @@ var latinLookalike = map[rune]rune{
 	'\u0410': 'A', '\u0412': 'B', '\u0415': 'E', '\u041a': 'K', '\u041c': 'M', '\u041d': 'H', '\u041e': 'O', '\u0420': 'P', '\u0421': 'C', '\u0422': 'T', '\u0425': 'X',
 	'\u03bf': 'o', '\u03bd': 'v', '\u03c1': 'p', '\u0391': 'A', '\u0392': 'B', '\u0395': 'E', '\u039f': 'O', '\u03a1': 'P',
 }
+
+// WrapNames turns a bare list of client names (a regexp alternation such as
+// "acme|globex") into a pattern that matches a name only as a whole word, with
+// word edges decided by Unicode letters and digits: Go's \b is ASCII-only, so
+// with it a name that begins or ends in a letter like Å, é or Ø would never
+// match. An underscore or punctuation is a boundary (trucker_ecommerce matches
+// "trucker"), a letter or digit glued on is not (acmecorp, acme2).
+//
+// The list is validated first, on its own: it must compile as a regexp, must not
+// match the empty string (an empty alternative such as "a||b" would match
+// everywhere), and must be self-contained: a value like "x)|.*(" compiles alone
+// only if it is balanced, so it cannot break out of the group it is put in.
+func WrapNames(names string) (string, error) {
+	if strings.TrimSpace(names) == "" {
+		return "", fmt.Errorf("client names list is empty")
+	}
+	re, err := regexp.Compile("(?i)" + names)
+	if err != nil {
+		return "", fmt.Errorf("client names are not a valid regexp alternation: %w", err)
+	}
+	if re.MatchString("") {
+		return "", fmt.Errorf("client names contain an empty alternative; it would match everywhere")
+	}
+	return `(?:^|[^\p{L}\p{N}])(?:` + names + `)(?:[^\p{L}\p{N}]|$)`, nil
+}

ingestion/internal/wing/router_test.go

@@ -234,3 +234,59 @@ func TestRouter_ClientMarkerChunkingDoesNotSplitAMultiByteRune(t *testing.T) {
 	clean := strings.Repeat("\u00e9", 300)
 	assert.Equal(t, wing.Default, r.Resolve(context.Background(), wing.Note{Text: clean}).Source, "no panic or false match on a text of 2-byte runes")
 }
+
+// From the review of #214: the shared secret is a bare list of client names.
+func TestWrapNames(t *testing.T) {
+	park := func(t *testing.T, names, text string) bool {
+		t.Helper()
+		pat, err := wing.WrapNames(names)
+		assert.NoError(t, err)
+		r, err := wing.NewRouter(routerBrain(t), "web-clip", pat)
+		assert.NoError(t, err)
+		return r.Resolve(context.Background(), wing.Note{Text: text}).Source == wing.Exception
+	}
+	t.Run("a short name does not match inside ordinary words", func(t *testing.T) {
+		assert.False(t, park(t, "dif|trucker", "a different approach to modifying prompts"))
+		assert.True(t, park(t, "dif|trucker", "kickoff for the DIF project"))
+	})
+	t.Run("names with a non-ASCII letter at the edge still match (Go's \\b would never)", func(t *testing.T) {
+		assert.True(t, park(t, "Åke|Société|Ørsted", "Åke called"))
+		assert.True(t, park(t, "Åke|Société|Ørsted", "la Société wrote"))
+		assert.True(t, park(t, "Åke|Société|Ørsted", "Ørsted wind"))
+	})
+	t.Run("identifiers with underscores and punctuation count as boundaries", func(t *testing.T) {
+		assert.True(t, park(t, "trucker", "repo trucker_ecommerce was cloned"))
+		assert.True(t, park(t, "trucker", "(Trucker)"))
+	})
+	t.Run("a name inside a longer word or with digits glued on is not a boundary", func(t *testing.T) {
+		assert.False(t, park(t, "acme", "acmecorp"))
+		assert.False(t, park(t, "acme", "acme2"))
+	})
+}
+
+func TestWrapNames_RefusesWhatWouldBreakOrBlockEverything(t *testing.T) {
+	for _, bad := range []string{
+		"x)|.*(",     // breaks out of the group
+		"acme)|(b",   // changes the structure
+		"a||b",       // empty alternative matches everywhere
+		"acme|",      // trailing empty alternative
+		"|acme",      // leading empty alternative
+		"",           // nothing
+		"c++ corp",   // not a valid regexp
+		"([unclosed", // not a valid regexp
+	} {
+		t.Run(bad, func(t *testing.T) {
+			_, err := wing.WrapNames(bad)
+			assert.Error(t, err)
+		})
+	}
+}
+
+func TestNewRouter_RefusesAPatternThatMatchesTheEmptyString(t *testing.T) {
+	for _, p := range []string{`.*`, `x*`, `^`, `\b`, `(a|)`} {
+		t.Run(p, func(t *testing.T) {
+			_, err := wing.NewRouter(routerBrain(t), "web-clip", p)
+			assert.Error(t, err, "it would park every note")
+		})
+	}
+}