feat: bind evidence credentials through local files

harness/tests/test_config_resources.py

@@ -384,3 +384,20 @@ evidence:
     legacy_source = build_evidence_sources(load_config(legacy))[0]
     assert isinstance(legacy_source, FilesystemEvidenceSource)
     assert legacy_source.root == (tmp_path / "curated").resolve()
+
+
+
+def test_safe_validation_formatter_keeps_location_and_type_without_rejected_input(tmp_path):
+    workspace = tmp_path / "workspace.yaml"
+    workspace.write_text("""
+dwh:
+  type: postgres_direct
+  connection: {database: analytics, schema: public, user: reader}
+""")
+
+    with pytest.raises(ConfigError) as caught:
+        load_config(workspace)
+
+    message = str(caught.value)
+    assert "dwh.postgres_direct.connection.password" in message
+    assert "missing" in message

harness/tests/test_registry_evidence_config.py

@@ -0,0 +1,280 @@
+import json
+
+import pytest
+import yaml
+from pydantic import SecretStr
+from typer.testing import CliRunner
+
+from tht.adapters.evidence import HttpManifestEvidenceSource
+from tht.adapters.factory import build_evidence_sources
+from tht.cli import app
+from tht.config import ConfigError, load_config
+
+SIGNED_CANARY = "SIGNED-CANARY-QUERY"
+ACCESS_CANARY = "ACCESS-CANARY"
+SECRET_CANARY = "SECRET-CANARY"
+TOKEN_CANARY = "TOKEN-CANARY"
+ALL_CANARIES = (SIGNED_CANARY, ACCESS_CANARY, SECRET_CANARY, TOKEN_CANARY)
+
+
+def raw_runtime(source, *, vector=None):
+    value = {
+        "dwh": {
+            "type": "postgres_direct",
+            "connection": {
+                "database": "analytics", "schema": "public", "user": "reader",
+                "password": "not-a-canary",
+            },
+        },
+        "evidence": {"sources": [source]},
+    }
+    if vector is not None:
+        value["vector"] = vector
+    return value
+
+
+def write_config(tmp_path, source, *, vector=None):
+    path = tmp_path / "runtime.yaml"
+    path.write_text(yaml.safe_dump(raw_runtime(source, vector=vector)))
+    return path
+
+
+def assert_no_canaries(value):
+    text = str(value)
+    for canary in ALL_CANARIES:
+        assert canary not in text
+
+
+def test_filesystem_config_does_not_touch_a_declared_source_root(tmp_path):
+    missing = tmp_path / "deliberately-missing"
+    cfg = load_config(write_config(tmp_path, {"type": "filesystem", "root": str(missing)}))
+
+    assert cfg.evidence.sources[0].root == missing
+    assert cfg.evidence.sources[0].patterns == ["**/*.md"]
+    assert cfg.evidence.sources[0].max_bytes == 10 * 1024 * 1024
+    assert not missing.exists()
+
+
+def test_public_http_urls_are_secret_typed_without_adapter_construction(tmp_path):
+    cfg = load_config(write_config(tmp_path, {
+        "type": "http", "urls": ["https://evidence.example.test/guide.md"],
+    }))
+    source = cfg.evidence.sources[0]
+
+    assert isinstance(source.urls[0], SecretStr)
+    assert source.transport_urls() == ["https://evidence.example.test/guide.md"]
+    assert source.connect_timeout == 5
+    assert source.read_timeout == 30
+    assert source.max_bytes == 10 * 1024 * 1024
+    assert source.max_redirects == 5
+    assert source.allow_private_hosts is False
+    assert source.max_cache_bytes == 64 * 1024 * 1024
+
+
+def test_signed_http_file_resolves_in_memory_and_preserves_provenance_order(tmp_path):
+    signed = [
+        f"https://evidence.example.test/guide.md?token={SIGNED_CANARY}",
+        "https://evidence.example.test/runbook.md?signature=second",
+    ]
+    secret_file = tmp_path / "signed-urls.json"
+    secret_file.write_text(json.dumps(signed))
+    cfg = load_config(write_config(tmp_path, {
+        "type": "http",
+        "provenance_urls": [
+            "https://evidence.example.test/guide.md",
+            "https://evidence.example.test/runbook.md",
+        ],
+        "signed_urls_file": str(secret_file),
+        "connect_timeout": 7,
+        "read_timeout": 41,
+        "max_bytes": 1234,
+        "max_redirects": 2,
+        "allow_private_hosts": True,
+        "max_cache_bytes": 5678,
+    }))
+    source = cfg.evidence.sources[0]
+
+    assert all(isinstance(url, SecretStr) for url in source.urls)
+    assert source.transport_urls() == signed
+    assert source.provenance_urls == [
+        "https://evidence.example.test/guide.md",
+        "https://evidence.example.test/runbook.md",
+    ]
+    assert (source.connect_timeout, source.read_timeout) == (7, 41)
+    assert (source.max_bytes, source.max_redirects, source.max_cache_bytes) == (1234, 2, 5678)
+    assert source.allow_private_hosts is True
+    assert "signed_urls_file" not in repr(source)
+    assert_no_canaries(repr(cfg))
+    assert_no_canaries(cfg.model_dump_json())
+
+    adapter = build_evidence_sources(cfg)[0]
+    assert isinstance(adapter, HttpManifestEvidenceSource)
+    assert_no_canaries(repr(adapter))
+
+
+@pytest.mark.parametrize("contents", [
+    "{malformed", json.dumps({"url": "https://evidence.example.test/guide.md"}),
+    json.dumps([]), json.dumps(["https://evidence.example.test/guide.md", 3]),
+])
+def test_signed_http_rejects_malformed_non_list_empty_or_non_string_files(tmp_path, contents):
+    secret_file = tmp_path / "signed-urls.json"
+    secret_file.write_text(contents)
+    path = write_config(tmp_path, {
+        "type": "http",
+        "provenance_urls": ["https://evidence.example.test/guide.md"],
+        "signed_urls_file": str(secret_file),
+    })
+
+    with pytest.raises(ConfigError) as caught:
+        load_config(path)
+    assert "signed URL file" in str(caught.value)
+    assert_no_canaries(caught.value)
+
+
+def test_signed_http_rejects_missing_and_oversized_files_without_disclosure(tmp_path):
+    missing = tmp_path / "missing.json"
+    path = write_config(tmp_path, {
+        "type": "http",
+        "provenance_urls": ["https://evidence.example.test/guide.md"],
+        "signed_urls_file": str(missing),
+    })
+    with pytest.raises(ConfigError, match="signed URL file"):
+        load_config(path)
+
+    oversized = tmp_path / "oversized.json"
+    oversized.write_bytes(b"x" * (1024 * 1024 + 1))
+    path = write_config(tmp_path, {
+        "type": "http",
+        "provenance_urls": ["https://evidence.example.test/guide.md"],
+        "signed_urls_file": str(oversized),
+    })
+    with pytest.raises(ConfigError, match="signed URL file") as caught:
+        load_config(path)
+    assert_no_canaries(caught.value)
+
+
+@pytest.mark.parametrize("provenance,signed", [
+    (
+        ["https://evidence.example.test/a.md", "https://evidence.example.test/b.md"],
+        ["https://evidence.example.test/b.md?sig=1", "https://evidence.example.test/a.md?sig=2"],
+    ),
+    (["https://evidence.example.test/a.md"], [
+        "https://evidence.example.test/a.md?sig=1", "https://evidence.example.test/b.md?sig=2",
+    ]),
+    (["https://evidence.example.test/a.md"], ["https://evidence.example.test/b.md"]),
+    (["https://evidence.example.test/a.md"], [f"https://user:{SIGNED_CANARY}@evidence.example.test/a.md"]),
+    (
+        ["https://evidence.example.test/a.md", "https://evidence.example.test/a.md"],
+        ["https://evidence.example.test/a.md?sig=1", "https://evidence.example.test/a.md?sig=2"],
+    ),
+])
+def test_signed_http_rejects_reordered_extra_mismatch_userinfo_and_duplicate_provenance(
+    tmp_path, provenance, signed,
+):
+    secret_file = tmp_path / "signed-urls.json"
+    secret_file.write_text(json.dumps(signed))
+    path = write_config(tmp_path, {
+        "type": "http", "provenance_urls": provenance, "signed_urls_file": str(secret_file),
+    })
+
+    with pytest.raises(ConfigError) as caught:
+        load_config(path)
+    assert "evidence.sources.0" in str(caught.value)
+    assert_no_canaries(caught.value)
+
+
+def test_s3_ambient_and_static_file_credentials_are_secret_typed(tmp_path):
+    ambient = load_config(write_config(tmp_path, {
+        "type": "s3", "bucket": "clinical-evidence", "prefix": "published/",
+    })).evidence.sources[0]
+    assert ambient.access_key is None
+    assert ambient.secret_key is None
+    assert ambient.session_token is None
+    assert ambient.max_bytes == 10 * 1024 * 1024
+    assert ambient.max_objects == 10_000
+    assert ambient.max_pages == 100
+    assert ambient.page_size == 1000
+
+    files = {}
+    for name, canary in [
+        ("access_key", ACCESS_CANARY), ("secret_key", SECRET_CANARY),
+        ("session_token", TOKEN_CANARY),
+    ]:
+        path = tmp_path / name
+        path.write_text(canary)
+        files[f"{name}_file"] = str(path)
+    cfg = load_config(write_config(tmp_path, {
+        "type": "s3", "bucket": "clinical-evidence", "prefix": "published/",
+        **files,
+        "endpoint_url": "https://s3.example.test",
+        "region": "eu-west-1",
+        "trusted_endpoint": True,
+        "allow_private_endpoint": True,
+        "allow_insecure_endpoint": False,
+        "max_bytes": 222,
+        "max_objects": 33,
+        "max_pages": 4,
+        "page_size": 5,
+    }))
+    source = cfg.evidence.sources[0]
+    assert all(isinstance(value, SecretStr) for value in (
+        source.access_key, source.secret_key, source.session_token,
+    ))
+    assert (source.max_bytes, source.max_objects, source.max_pages, source.page_size) == (222, 33, 4, 5)
+    assert_no_canaries(repr(cfg))
+    assert_no_canaries(cfg.model_dump_json())
+
+
+def test_evidence_policy_defaults_non_defaults_and_unknown_keys(tmp_path):
+    default = load_config(write_config(tmp_path, {
+        "type": "filesystem", "root": str(tmp_path / "missing"),
+    }))
+    assert default.vector.max_chunk_chars == 4000
+    assert default.vector.retain_published_generations == 3
+
+    explicit = load_config(write_config(tmp_path, {
+        "type": "filesystem", "root": str(tmp_path / "missing"),
+        "patterns": ["docs/*.md"], "max_bytes": 99,
+    }, vector={"max_chunk_chars": 123, "retain_published_generations": 7}))
+    assert explicit.evidence.sources[0].patterns == ["docs/*.md"]
+    assert explicit.vector.max_chunk_chars == 123
+    assert explicit.vector.retain_published_generations == 7
+
+    for mutation in [
+        {"type": "filesystem", "root": str(tmp_path), "unknown": SIGNED_CANARY},
+        {"type": "http", "urls": ["https://evidence.example.test/a"], "unknown": SIGNED_CANARY},
+        {"type": "s3", "bucket": "bucket-name", "unknown": SIGNED_CANARY},
+    ]:
+        with pytest.raises(ConfigError) as caught:
+            load_config(write_config(tmp_path, mutation))
+        assert "extra_forbidden" in str(caught.value)
+        assert_no_canaries(caught.value)
+
+
+def test_validation_repr_cli_and_exception_output_never_disclose_transport_secrets(tmp_path):
+    secret_file = tmp_path / "signed-urls.json"
+    secret_file.write_text(json.dumps([
+        f"https://evidence.example.test/other.md?token={SIGNED_CANARY}",
+    ]))
+    path = write_config(tmp_path, {
+        "type": "http",
+        "provenance_urls": ["https://evidence.example.test/guide.md"],
+        "signed_urls_file": str(secret_file),
+    })
+    with pytest.raises(ConfigError) as caught:
+        load_config(path)
+    assert_no_canaries(caught.value)
+
+    valid_file = tmp_path / "valid-signed-urls.json"
+    valid_file.write_text(json.dumps([
+        f"https://evidence.example.test/guide.md?token={SIGNED_CANARY}",
+    ]))
+    valid = write_config(tmp_path, {
+        "type": "http",
+        "provenance_urls": ["https://evidence.example.test/guide.md"],
+        "signed_urls_file": str(valid_file),
+    })
+    result = CliRunner().invoke(app, ["config", "check", "--config", str(valid)])
+    assert result.exit_code == 0
+    assert_no_canaries(result.stdout)
+    assert_no_canaries(result.stderr)

harness/tht/adapters/factory.py

@@ -64,7 +64,7 @@ def build_evidence_sources(cfg: Config):
             case "http":
                 sources.append(
                     HttpManifestEvidenceSource(
-                        [url.get_secret_value() for url in resource.urls],
+                        resource.transport_urls(),
                         connect_timeout=resource.connect_timeout,
                         read_timeout=resource.read_timeout,
                         max_bytes=resource.max_bytes,

harness/tht/config.py

@@ -1,5 +1,7 @@
+import json
 import os
 import re
+import stat
 import warnings
 from ipaddress import ip_address
 from pathlib import Path
@@ -10,6 +12,7 @@ import yaml
 from pydantic import BaseModel, Field, PrivateAttr, SecretStr, ValidationError, model_validator
 
 from tht.config_compat import translate_legacy_config
+from tht.ports.evidence import canonical_provenance_uri
 
 _ENV_RE = re.compile(r"\$\{([A-Za-z_][A-Za-z0-9_]*)\}")
 
@@ -43,6 +46,51 @@ def _expand_env(value: Any) -> Any:
     return value
 
 
+_MAX_SIGNED_URL_FILE_BYTES = 1024 * 1024
+
+
+def _resolve_http_signed_url_files(value: Any) -> Any:
+    """Resolve only signed HTTP URL arrays, keeping their values out of public errors."""
+    if isinstance(value, dict):
+        resolved = {
+            key: _resolve_http_signed_url_files(item)
+            for key, item in value.items()
+        }
+        if resolved.get("type") != "http" or "signed_urls_file" not in resolved:
+            return resolved
+        if "urls" in resolved:
+            raise ConfigError("HTTP signed URL file cannot be combined with urls")
+        path_value = resolved.pop("signed_urls_file")
+        if not isinstance(path_value, str):
+            raise ConfigError("Invalid signed URL file reference")
+        path = Path(path_value)
+        try:
+            entry = path.lstat()
+            target = path.stat()
+            if stat.S_ISLNK(entry.st_mode) or not stat.S_ISREG(target.st_mode):
+                raise OSError
+            if target.st_size > _MAX_SIGNED_URL_FILE_BYTES:
+                raise OSError
+            with path.open("rb") as stream:
+                payload = stream.read(_MAX_SIGNED_URL_FILE_BYTES + 1)
+            if len(payload) > _MAX_SIGNED_URL_FILE_BYTES:
+                raise OSError
+            parsed = json.loads(payload.decode("utf-8"))
+        except (OSError, UnicodeError, json.JSONDecodeError) as exc:
+            raise ConfigError("Cannot read signed URL file") from exc
+        if (
+            not isinstance(parsed, list)
+            or not parsed
+            or any(not isinstance(item, str) or not item for item in parsed)
+        ):
+            raise ConfigError("Invalid signed URL file")
+        resolved["urls"] = parsed
+        return resolved
+    if isinstance(value, list):
+        return [_resolve_http_signed_url_files(item) for item in value]
+    return value
+
+
 def _resolve_secret_files(value: Any) -> Any:
     if isinstance(value, dict):
         resolved = {key: _resolve_secret_files(item) for key, item in value.items()}
@@ -228,12 +276,15 @@ class FilesystemEvidenceSourceConfig(BaseModel):
     patterns: list[str] = ["**/*.md"]
     max_bytes: int = Field(default=10 * 1024 * 1024, gt=0)
 
+    model_config = {"extra": "forbid"}
+
 
 class HttpEvidenceSourceConfig(BaseModel):
     type: Literal["http"]
-    # Manifest URLs may contain signed query parameters. Treat the complete transport URL as
-    # secret-bearing configuration; adapters derive a query-free provenance URI from it.
+    # Transport URLs are secret-bearing. Signed-file configurations retain only public,
+    # query-free provenance identities alongside the masked transport values.
     urls: list[SecretStr] = Field(min_length=1)
+    provenance_urls: list[str] | None = Field(default=None, min_length=1)
     connect_timeout: float = Field(default=5, gt=0)
     read_timeout: float = Field(default=30, gt=0)
     max_bytes: int = Field(default=10 * 1024 * 1024, gt=0)
@@ -241,6 +292,38 @@ class HttpEvidenceSourceConfig(BaseModel):
     allow_private_hosts: bool = False
     max_cache_bytes: int = Field(default=64 * 1024 * 1024, gt=0)
 
+    model_config = {"extra": "forbid"}
+
+    @model_validator(mode="after")
+    def validate_provenance_mapping(self):
+        transport_urls = [url.get_secret_value() for url in self.urls]
+        try:
+            canonical = [canonical_provenance_uri(url) for url in transport_urls]
+        except ValueError as exc:
+            raise ValueError("HTTP transport URL is invalid") from exc
+        if any(
+            urlparse(url).scheme not in ("http", "https") or not urlparse(url).hostname
+            for url in transport_urls
+        ):
+            raise ValueError("HTTP transport URL must use http or https")
+        if self.provenance_urls is None:
+            return self
+        try:
+            provenance = [canonical_provenance_uri(url) for url in self.provenance_urls]
+        except ValueError as exc:
+            raise ValueError("HTTP provenance URL is invalid") from exc
+        if provenance != self.provenance_urls:
+            raise ValueError("HTTP provenance URLs must be canonical query-free identities")
+        if len(set(provenance)) != len(provenance):
+            raise ValueError("HTTP provenance URLs must not repeat")
+        if len(canonical) != len(provenance) or canonical != provenance:
+            raise ValueError("Signed HTTP URLs must map one-to-one to provenance URLs in order")
+        return self
+
+    def transport_urls(self) -> list[str]:
+        """Expose secret transport values only at the adapter-construction boundary."""
+        return [url.get_secret_value() for url in self.urls]
+
 
 class S3EvidenceSourceConfig(BaseModel):
     type: Literal["s3"]
@@ -259,6 +342,16 @@ class S3EvidenceSourceConfig(BaseModel):
     max_pages: int = Field(default=100, gt=0)
     page_size: int = Field(default=1000, gt=0, le=1000)
 
+    model_config = {"extra": "forbid"}
+
+    @model_validator(mode="after")
+    def validate_static_credentials(self):
+        if (self.access_key is None) != (self.secret_key is None):
+            raise ValueError("S3 access_key and secret_key must be configured together")
+        if self.session_token is not None and self.access_key is None:
+            raise ValueError("S3 session_token requires static credentials")
+        return self
+
 
 EvidenceSourceConfig = Annotated[
     FilesystemEvidenceSourceConfig | HttpEvidenceSourceConfig | S3EvidenceSourceConfig,
@@ -282,6 +375,8 @@ class EvidenceSourcesConfig(BaseModel):
             raise ValueError("evidence requires source_root or sources")
         return self
 
+    model_config = {"extra": "forbid"}
+
 
 class EmbeddingsConfig(BaseModel):
     provider: str = "ollama_internal"
@@ -298,10 +393,12 @@ class EmbeddingsConfig(BaseModel):
 
 
 class VectorConfig(BaseModel):
-    max_chunk_chars: int = 4000
+    max_chunk_chars: int = Field(default=4000, gt=0)
     # ACTIVE plus the two most recent rollback generations by default.
     retain_published_generations: int = Field(default=3, ge=1)
 
+    model_config = {"extra": "forbid"}
+
 
 class SearchConfig(BaseModel):
     rrf_k: int = 60
@@ -392,6 +489,31 @@ def workspace_id_for_config(config: Config, path: Path) -> str:
     return workspace_id_from_path(path)
 
 
+def _format_validation_error(error: ValidationError) -> str:
+    messages = {
+        "missing": "required field",
+        "extra_forbidden": "unknown field",
+        "greater_than": "value must be greater than the configured bound",
+        "greater_than_equal": "value must meet the configured lower bound",
+        "less_than_equal": "value exceeds the configured upper bound",
+        "literal_error": "unsupported literal value",
+        "union_tag_invalid": "unsupported discriminator",
+        "union_tag_not_found": "missing discriminator",
+        "string_too_short": "string is too short",
+        "too_short": "collection is too short",
+        "value_error": "configuration value is invalid",
+    }
+    lines = []
+    for issue in error.errors(include_input=False, include_url=False):
+        location = ".".join(str(part) for part in issue.get("loc", ())) or "configuration"
+        error_type = str(issue.get("type", "validation_error"))
+        message = messages.get(error_type, "invalid configuration value")
+        if location == "runtime_identity" and error_type == "value_error":
+            message = "source_identity does not match workspace identity"
+        lines.append(f"{location} [{error_type}]: {message}")
+    return "\n".join(lines)
+
+
 def load_config(path: Path) -> Config:
     if not path.exists():
         raise ConfigError(f"File di configurazione non trovato: {path}")
@@ -401,7 +523,7 @@ def load_config(path: Path) -> Config:
         raise ConfigError(f"Configurazione YAML non valida: {path}") from exc
     if not isinstance(raw, dict):
         raise ConfigError(f"Configurazione non valida (atteso un mapping YAML): {path}")
-    expanded = _resolve_secret_files(_expand_env(raw))
+    expanded = _resolve_secret_files(_resolve_http_signed_url_files(_expand_env(raw)))
     _validate_internal_embedding_contract(expanded, path)
     _validate_internal_vector_contract(expanded, path)
     translated, used_legacy = translate_legacy_config(expanded)
@@ -409,7 +531,8 @@ def load_config(path: Path) -> Config:
     try:
         cfg = Config.model_validate(translated)
     except ValidationError as e:
-        raise ConfigError(f"Configurazione non valida in {path}:\n{e}") from e
+        details = _format_validation_error(e)
+        raise ConfigError(f"Configurazione non valida in {path}:\n{details}") from e
     env_profile = os.environ.get("THT_PROFILE")
     if env_profile is not None:
         if env_profile not in ("server", "workstation"):