claude-dev-env 2.4.0 → 2.5.0

package/skills/plan-to-pr/scripts/validate_protocol.py

@@ -0,0 +1,622 @@
+"""Validate one skill protocol task-run record."""
+
+from __future__ import annotations
+
+import json
+import hashlib
+import re
+import subprocess
+import sys
+from collections.abc import Mapping, Sequence
+from pathlib import Path
+
+from config.constants import (
+    ALL_ALLOWED_RECORD_FIELDS,
+    ALL_RECORD_FIELDS,
+    ALL_REPAIR_FIELDS,
+    ALL_REPAIR_TEXT_FIELDS,
+    ALL_REVIEW_FIELDS,
+    ALL_REVIEW_TEXT_FIELDS,
+    ALL_TOP_LEVEL_TEXT_FIELDS,
+    ALL_SUPPORTED_SCHEMA_TYPES,
+    ALL_VERIFICATION_FIELDS,
+    ALLOWED_REVIEW_MODEL,
+    ALLOWED_WORKER_EFFORT,
+    ALLOWED_WORKER_ROLE,
+    ARGUMENT_COUNT_REQUIRED,
+    ARGUMENT_COUNT_WITH_WORKTREE,
+    OPTION_ARGUMENT_INDEX,
+    WORKTREE_OPTION,
+    WORKTREE_PATH_ARGUMENT_INDEX,
+    COMMIT_HASH_ERROR_TEMPLATE,
+    COMMIT_HASH_PATTERN,
+    EXIT_CODE_INVALID_RECORD,
+    EXIT_CODE_VALID_RECORD,
+    JSON_ENCODING,
+    MISSING_FIELD_SEPARATOR,
+    MAXIMUM_COMMIT_HASH_LENGTH,
+    MINIMUM_COMMIT_HASH_LENGTH,
+    PROTOCOL_VALIDATION_PASSED,
+    RECORD_SCHEMA_FILENAME,
+    DOCS_PLANS_PATH_PATTERN,
+    FINDING_STATUS_CONFIRMED,
+    FINDING_STATUS_DISMISSED,
+    NATIVE_REVIEW_COMMAND,
+    PASSED_EVIDENCE_PREFIX,
+    REPAIR_STATUS_COMPLETE,
+    REPAIR_STATUS_NOT_REQUIRED,
+    SURFACE_HASH_PATTERN,
+    PATH_SEPARATOR_PATTERN,
+    SHA256_ALGORITHM,
+)
+
+
+class ProtocolValidationError(ValueError):
+    """Identify a task-run record contract violation."""
+
+
+def _load_schema_object(schema_path: Path) -> Mapping[str, object]:
+    """Load one JSON schema object."""
+    try:
+        parsed_schema = json.loads(schema_path.read_text(encoding=JSON_ENCODING))
+    except json.JSONDecodeError as error:
+        raise ProtocolValidationError("record schema is invalid JSON") from error
+    except OSError as error:
+        raise ProtocolValidationError(f"cannot read record schema: {error}") from error
+    if not isinstance(parsed_schema, dict):
+        raise ProtocolValidationError("record schema must be a JSON object")
+    return parsed_schema
+
+
+def _resolve_schema_reference(
+    all_schema_node: Mapping[str, object], all_schema_root: Mapping[str, object]
+) -> Mapping[str, object]:
+    """Resolve a local `$defs` schema reference."""
+    schema_reference = all_schema_node.get("$ref")
+    if schema_reference is None:
+        return all_schema_node
+    if not isinstance(schema_reference, str) or not schema_reference.startswith(
+        "#/$defs/"
+    ):
+        raise ProtocolValidationError("record schema has an unsupported reference")
+    definitions = all_schema_root.get("$defs")
+    if not isinstance(definitions, dict):
+        raise ProtocolValidationError("record schema has no definitions")
+    definition_name = schema_reference.removeprefix("#/$defs/")
+    definition = definitions.get(definition_name)
+    if not isinstance(definition, dict):
+        raise ProtocolValidationError(
+            f"record schema definition is missing: {definition_name}"
+        )
+    return definition
+
+
+def _validate_schema_node(
+    candidate: object,
+    all_schema_node: Mapping[str, object],
+    all_schema_root: Mapping[str, object],
+    field_path: str,
+) -> None:
+    """Validate one record value against the supported schema keywords."""
+    resolved_schema = _resolve_schema_reference(all_schema_node, all_schema_root)
+    if "const" in resolved_schema and candidate != resolved_schema["const"]:
+        raise ProtocolValidationError(
+            f"{field_path} does not match its schema constant"
+        )
+    schema_type = resolved_schema.get("type")
+    if schema_type == "object":
+        _validate_schema_object(candidate, resolved_schema, all_schema_root, field_path)
+        return
+    if schema_type == "array":
+        _validate_schema_array(candidate, resolved_schema, all_schema_root, field_path)
+        return
+    if schema_type == "string":
+        _validate_schema_string(candidate, resolved_schema, field_path)
+        return
+    if schema_type == "boolean" and not isinstance(candidate, bool):
+        raise ProtocolValidationError(f"{field_path} must be a boolean")
+    if schema_type not in (*ALL_SUPPORTED_SCHEMA_TYPES, None):
+        raise ProtocolValidationError(f"{field_path} has an unsupported schema type")
+
+
+def _validate_schema_object(
+    candidate: object,
+    all_schema_node: Mapping[str, object],
+    all_schema_root: Mapping[str, object],
+    field_path: str,
+) -> None:
+    """Validate an object schema node."""
+    if not isinstance(candidate, dict):
+        raise ProtocolValidationError(f"{field_path} must be an object")
+    required_fields = all_schema_node.get("required", [])
+    if not isinstance(required_fields, list):
+        raise ProtocolValidationError(f"{field_path} has invalid required fields")
+    for each_field_name in required_fields:
+        if isinstance(each_field_name, str) and each_field_name not in candidate:
+            raise ProtocolValidationError(f"{field_path} missing: {each_field_name}")
+    properties = all_schema_node.get("properties", {})
+    if not isinstance(properties, dict):
+        raise ProtocolValidationError(f"{field_path} has invalid properties")
+    if all_schema_node.get("additionalProperties") is False:
+        additional_fields = sorted(
+            each_field_name
+            for each_field_name in candidate
+            if each_field_name not in properties
+        )
+        if additional_fields:
+            raise ProtocolValidationError(
+                f"{field_path} has additional fields: "
+                f"{MISSING_FIELD_SEPARATOR.join(additional_fields)}"
+            )
+    for each_field_name, each_schema_node in properties.items():
+        if each_field_name in candidate and isinstance(each_schema_node, dict):
+            _validate_schema_node(
+                candidate[each_field_name],
+                each_schema_node,
+                all_schema_root,
+                f"{field_path}.{each_field_name}",
+            )
+
+
+def _validate_schema_array(
+    candidate: object,
+    all_schema_node: Mapping[str, object],
+    all_schema_root: Mapping[str, object],
+    field_path: str,
+) -> None:
+    """Validate an array schema node."""
+    if not isinstance(candidate, list):
+        raise ProtocolValidationError(f"{field_path} must be an array")
+    minimum_array_entries = all_schema_node.get("minItems", 0)
+    if (
+        isinstance(minimum_array_entries, int)
+        and len(candidate) < minimum_array_entries
+    ):
+        raise ProtocolValidationError(
+            f"{field_path} must contain at least {minimum_array_entries} items"
+        )
+    array_entry_schema = all_schema_node.get("items")
+    if isinstance(array_entry_schema, dict):
+        for each_index, each_candidate in enumerate(candidate):
+            _validate_schema_node(
+                each_candidate,
+                array_entry_schema,
+                all_schema_root,
+                f"{field_path}[{each_index}]",
+            )
+
+
+def _validate_schema_string(
+    candidate: object, all_schema_node: Mapping[str, object], field_path: str
+) -> None:
+    """Validate a string schema node."""
+    if not isinstance(candidate, str):
+        raise ProtocolValidationError(f"{field_path} must be a string")
+    minimum_length = all_schema_node.get("minLength", 0)
+    if isinstance(minimum_length, int) and len(candidate) < minimum_length:
+        raise ProtocolValidationError(f"{field_path} must not be empty")
+    pattern = all_schema_node.get("pattern")
+    if isinstance(pattern, str) and re.search(pattern, candidate) is None:
+        raise ProtocolValidationError(f"{field_path} does not match its schema pattern")
+
+
+def _load_json_object(task_run_path: Path) -> Mapping[str, object]:
+    """Load one task-run object from JSON.
+
+    Args:
+        task_run_path: JSON file containing one task-run record.
+
+    Returns:
+        The decoded task-run mapping.
+
+    Raises:
+        ProtocolValidationError: If the file cannot be read or decoded as an object.
+    """
+    try:
+        parsed_task_run = json.loads(task_run_path.read_text(encoding=JSON_ENCODING))
+    except json.JSONDecodeError as error:
+        raise ProtocolValidationError("invalid JSON") from error
+    except UnicodeDecodeError as error:
+        raise ProtocolValidationError("invalid UTF-8") from error
+    except OSError as error:
+        raise ProtocolValidationError(
+            f"cannot read task-run record: {error}"
+        ) from error
+    if not isinstance(parsed_task_run, dict):
+        raise ProtocolValidationError("task-run record must be a JSON object")
+    return parsed_task_run
+
+
+def _require_fields(
+    all_task_run_record: Mapping[str, object],
+    all_required_field_names: Sequence[str],
+    record_name: str,
+) -> None:
+    """Require named fields in a task-run mapping.
+
+    Args:
+        all_task_run_record: Mapping to inspect.
+        all_required_field_names: Field names that must be present.
+        record_name: Display name used in validation errors.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If a required field is absent.
+    """
+    missing_field_names = [
+        each_field_name
+        for each_field_name in all_required_field_names
+        if each_field_name not in all_task_run_record
+    ]
+    if missing_field_names:
+        raise ProtocolValidationError(
+            f"{record_name} missing: "
+            f"{MISSING_FIELD_SEPARATOR.join(missing_field_names)}"
+        )
+
+
+def _reject_additional_fields(
+    all_task_run_record: Mapping[str, object],
+    all_allowed_field_names: Sequence[str],
+    record_name: str,
+) -> None:
+    """Reject fields that are not named by the record schema.
+
+    Args:
+        all_task_run_record: Mapping to inspect.
+        all_allowed_field_names: Field names accepted by the schema.
+        record_name: Display name used in validation errors.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If an unknown field is present.
+    """
+    allowed_field_names = set(all_allowed_field_names)
+    additional_field_names = sorted(
+        each_field_name
+        for each_field_name in all_task_run_record
+        if each_field_name not in allowed_field_names
+    )
+    if additional_field_names:
+        raise ProtocolValidationError(
+            f"{record_name} has additional fields: "
+            f"{MISSING_FIELD_SEPARATOR.join(additional_field_names)}"
+        )
+
+
+def _require_text(
+    all_task_run_record: Mapping[str, object], field_name: str, record_name: str
+) -> None:
+    """Require a non-empty text field in a task-run mapping.
+
+    Args:
+        all_task_run_record: Mapping containing the field.
+        field_name: Name of the field to validate.
+        record_name: Display name used in validation errors.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If the field is absent, non-text, or empty.
+    """
+    field_text = all_task_run_record[field_name]
+    if not isinstance(field_text, str) or not field_text:
+        raise ProtocolValidationError(
+            f"{record_name}.{field_name} must be a non-empty string"
+        )
+
+
+def _require_string_list(
+    all_task_run_record: Mapping[str, object], field_name: str, record_name: str
+) -> list[str]:
+    """Require a list containing only non-empty strings.
+
+    Args:
+        all_task_run_record: Mapping containing the field.
+        field_name: Name of the field to validate.
+        record_name: Display name used in validation errors.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If the field is not a string list.
+    """
+    all_field_entries = all_task_run_record[field_name]
+    if not isinstance(all_field_entries, list) or any(
+        not isinstance(each_entry, str) or not each_entry
+        for each_entry in all_field_entries
+    ):
+        raise ProtocolValidationError(
+            f"{record_name}.{field_name} must be a string list"
+        )
+    return all_field_entries
+
+
+def _require_non_empty_string_list(
+    all_task_run_record: Mapping[str, object], field_name: str, record_name: str
+) -> None:
+    """Require a non-empty list containing only non-empty strings.
+
+    Args:
+        all_task_run_record: Mapping containing the field.
+        field_name: Name of the field to validate.
+        record_name: Display name used in validation errors.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If the field is not a non-empty string list.
+    """
+    all_field_entries = _require_string_list(
+        all_task_run_record, field_name, record_name
+    )
+    if not all_field_entries:
+        raise ProtocolValidationError(f"{record_name}.{field_name} must not be empty")
+
+
+def _require_record(
+    all_task_run_record: Mapping[str, object], field_name: str
+) -> Mapping[str, object]:
+    """Return a nested task-run mapping.
+
+    Args:
+        all_task_run_record: Mapping containing the nested record.
+        field_name: Name of the nested record field.
+
+    Returns:
+        The nested task-run mapping.
+
+    Raises:
+        ProtocolValidationError: If the nested field is not an object.
+    """
+    nested_task_run_record = all_task_run_record[field_name]
+    if not isinstance(nested_task_run_record, dict):
+        raise ProtocolValidationError(f"record.{field_name} must be an object")
+    return nested_task_run_record
+
+
+def _validate_review_record(all_review_record: Mapping[str, object]) -> None:
+    """Validate the findings-only review section.
+
+    Args:
+        all_review_record: Review mapping from the task-run record.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If the review section violates its contract.
+    """
+    _reject_additional_fields(all_review_record, ALL_REVIEW_FIELDS, "review_record")
+    _require_fields(all_review_record, ALL_REVIEW_FIELDS, "review_record")
+    for each_field_name in ALL_REVIEW_TEXT_FIELDS:
+        _require_text(all_review_record, each_field_name, "review_record")
+    findings = all_review_record["findings"]
+    if not isinstance(findings, list) or any(
+        not isinstance(each_finding, dict)
+        or not isinstance(each_finding.get("finding"), str)
+        or each_finding.get("disposition") not in {
+            FINDING_STATUS_CONFIRMED,
+            FINDING_STATUS_DISMISSED,
+        }
+        for each_finding in findings
+    ):
+        raise ProtocolValidationError("review_record.findings must be classified")
+    if all_review_record["findings_only"] is not True:
+        raise ProtocolValidationError("review_record.findings_only must be true")
+    if all_review_record["has_repair_flag"] is not False:
+        raise ProtocolValidationError("review_record.has_repair_flag must be false")
+
+
+def _validate_repair_record(all_repair_record: Mapping[str, object]) -> None:
+    """Validate the confirmed-findings repair section.
+
+    Args:
+        all_repair_record: Repair mapping from the task-run record.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If the repair section violates its contract.
+    """
+    _reject_additional_fields(all_repair_record, ALL_REPAIR_FIELDS, "repair_record")
+    _require_fields(all_repair_record, ALL_REPAIR_FIELDS, "repair_record")
+    for each_field_name in ALL_REPAIR_TEXT_FIELDS:
+        _require_text(all_repair_record, each_field_name, "repair_record")
+    _require_string_list(all_repair_record, "confirmed_findings", "repair_record")
+
+
+def _validate_verification_record(
+    all_verification_record: Mapping[str, object], record_name: str
+) -> None:
+    """Validate one verification evidence section.
+
+    Args:
+        all_verification_record: Verification mapping from the task-run record.
+        record_name: Display name used in validation errors.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If required evidence is missing or empty.
+    """
+    _reject_additional_fields(
+        all_verification_record, ALL_VERIFICATION_FIELDS, record_name
+    )
+    _require_fields(all_verification_record, ALL_VERIFICATION_FIELDS, record_name)
+    for each_field_name in ALL_VERIFICATION_FIELDS:
+        _require_text(all_verification_record, each_field_name, record_name)
+        field_text = all_verification_record[each_field_name]
+        if (
+            each_field_name != "surface_hash"
+            and isinstance(field_text, str)
+            and not field_text.lower().startswith(PASSED_EVIDENCE_PREFIX)
+        ):
+            raise ProtocolValidationError(
+                f"{record_name}.{each_field_name} must be explicit passed evidence"
+            )
+
+
+def _normalize_allowed_path(raw_path: str) -> str:
+    normalized_path = PATH_SEPARATOR_PATTERN.sub("/", raw_path.strip())
+    normalized_path = normalized_path.removeprefix("./")
+    if not normalized_path or normalized_path.startswith("../") or "/../" in normalized_path:
+        raise ProtocolValidationError("allowed_files contains an unsafe path")
+    if normalized_path.startswith("/") or (
+        normalized_path[1:].startswith(":/")
+    ):
+        raise ProtocolValidationError("allowed_files contains an absolute path")
+    if DOCS_PLANS_PATH_PATTERN.match(normalized_path.lower()):
+        raise ProtocolValidationError("allowed_files cannot contain docs/plans paths")
+    return normalized_path
+
+
+def _validate_semantics(all_task_run_record: Mapping[str, object]) -> None:
+    worker_route = all_task_run_record["worker_route"]
+    if worker_route != f"{ALLOWED_WORKER_ROLE}; effort={ALLOWED_WORKER_EFFORT}":
+        raise ProtocolValidationError("worker_route must be the fast low-effort Luna contract")
+    all_allowed_files = all_task_run_record["allowed_files"]
+    assert isinstance(all_allowed_files, list)
+    normalized_files = [_normalize_allowed_path(each_path) for each_path in all_allowed_files if isinstance(each_path, str)]
+    if len(normalized_files) != len(set(normalized_files)):
+        raise ProtocolValidationError("allowed_files contains duplicate paths")
+    review_record = _require_record(all_task_run_record, "review_record")
+    if review_record["resolved_model"] != ALLOWED_REVIEW_MODEL or review_record["effort"] != ALLOWED_WORKER_EFFORT:
+        raise ProtocolValidationError("review_record must use fast low-effort Luna")
+    command = review_record["command"]
+    if not isinstance(command, str):
+        raise ProtocolValidationError("review_record.command must be a string")
+    if command != NATIVE_REVIEW_COMMAND or "--fix" in command:
+        raise ProtocolValidationError("review_record.command must be exactly /e-code-review low")
+    findings = review_record["findings"]
+    assert isinstance(findings, list)
+    confirmed_findings = [each_finding["finding"] for each_finding in findings if each_finding["disposition"] == FINDING_STATUS_CONFIRMED]
+    repair_record = _require_record(all_task_run_record, "repair_record")
+    if repair_record["confirmed_findings"] != confirmed_findings:
+        raise ProtocolValidationError("repair_record must list exactly the confirmed findings")
+    if confirmed_findings and repair_record["repair_status"] != REPAIR_STATUS_COMPLETE:
+        raise ProtocolValidationError("confirmed findings require a completed repair")
+    if not confirmed_findings and repair_record["repair_status"] != REPAIR_STATUS_NOT_REQUIRED:
+        raise ProtocolValidationError("repair is permitted only when no findings are confirmed")
+    hashes = [
+        review_record["surface_hash"], repair_record["surface_hash"],
+        _require_record(all_task_run_record, "reverification_record")["surface_hash"],
+        _require_record(all_task_run_record, "verification_record")["surface_hash"],
+    ]
+    if any(not isinstance(each_hash, str) or SURFACE_HASH_PATTERN.fullmatch(each_hash) is None for each_hash in hashes) or len(set(hashes)) != 1:
+        raise ProtocolValidationError("review and verification surface hashes must match")
+
+
+def _committed_surface_hash(worktree: Path, commit_hash: str, all_allowed_files: list[str]) -> str:
+    try:
+        completed = subprocess.run(
+            ["git", "diff", f"{commit_hash}^", commit_hash, "--", *all_allowed_files],
+            cwd=worktree, check=True, capture_output=True,
+        )
+    except (OSError, subprocess.CalledProcessError) as error:
+        raise ProtocolValidationError("cannot compute committed task surface") from error
+    return hashlib.new(SHA256_ALGORITHM, completed.stdout).hexdigest()
+
+
+def validate_record(
+    all_task_run_record: Mapping[str, object], schema_path: Path
+) -> None:
+    """Validate a complete task-run record against its JSON schema file.
+
+    Args:
+        all_task_run_record: Task-run mapping to validate.
+        schema_path: Path to the task-run schema JSON file.
+
+    Returns:
+        None.
+
+    Raises:
+        ProtocolValidationError: If the schema or task-run record is invalid.
+    """
+    schema_object = _load_schema_object(schema_path)
+    _validate_schema_node(all_task_run_record, schema_object, schema_object, "record")
+    _reject_additional_fields(all_task_run_record, ALL_ALLOWED_RECORD_FIELDS, "record")
+    _require_fields(all_task_run_record, ALL_RECORD_FIELDS, "record")
+    for each_field_name in ALL_TOP_LEVEL_TEXT_FIELDS:
+        _require_text(all_task_run_record, each_field_name, "record")
+    _require_non_empty_string_list(all_task_run_record, "allowed_files", "record")
+    commit_hash = all_task_run_record["commit"]
+    if (
+        not isinstance(commit_hash, str)
+        or not MINIMUM_COMMIT_HASH_LENGTH
+        <= len(commit_hash)
+        <= MAXIMUM_COMMIT_HASH_LENGTH
+        or COMMIT_HASH_PATTERN.fullmatch(commit_hash) is None
+    ):
+        raise ProtocolValidationError(
+            COMMIT_HASH_ERROR_TEMPLATE
+            % (MINIMUM_COMMIT_HASH_LENGTH, MAXIMUM_COMMIT_HASH_LENGTH)
+        )
+    _validate_review_record(_require_record(all_task_run_record, "review_record"))
+    _validate_repair_record(_require_record(all_task_run_record, "repair_record"))
+    _validate_verification_record(
+        _require_record(all_task_run_record, "reverification_record"),
+        "reverification_record",
+    )
+    _validate_verification_record(
+        _require_record(all_task_run_record, "verification_record"),
+        "verification_record",
+    )
+    _validate_semantics(all_task_run_record)
+    worktree = all_task_run_record.get("worktree")
+    if isinstance(worktree, str):
+        all_allowed_files = [
+            _normalize_allowed_path(each_path)
+            for each_path in _require_string_list(
+                all_task_run_record, "allowed_files", "record"
+            )
+        ]
+        committed_hash = _committed_surface_hash(Path(worktree), str(all_task_run_record["commit"]), all_allowed_files)
+        recorded_hash = _require_record(all_task_run_record, "verification_record")["surface_hash"]
+        if committed_hash != recorded_hash:
+            raise ProtocolValidationError("verification surface hash does not match the committed diff")
+
+
+def main(all_cli_arguments: Sequence[str]) -> int:
+    """Validate a task-run record from command-line arguments.
+
+    Args:
+        all_cli_arguments: Program name followed by one record path.
+
+    Returns:
+        Exit code `0` for a valid record or `2` for invalid input.
+    """
+    if len(all_cli_arguments) not in {
+        ARGUMENT_COUNT_REQUIRED,
+        ARGUMENT_COUNT_WITH_WORKTREE,
+    } or (
+        len(all_cli_arguments) == ARGUMENT_COUNT_WITH_WORKTREE
+        and all_cli_arguments[OPTION_ARGUMENT_INDEX] != WORKTREE_OPTION
+    ):
+        print("usage: validate_protocol.py <record.json> [--worktree PATH]", file=sys.stderr)
+        return EXIT_CODE_INVALID_RECORD
+    schema_path = Path(__file__).parent.parent / "reference" / RECORD_SCHEMA_FILENAME
+    try:
+        record = _load_json_object(Path(all_cli_arguments[1]))
+        if len(all_cli_arguments) == ARGUMENT_COUNT_WITH_WORKTREE:
+            record = dict(record)
+            record["worktree"] = all_cli_arguments[WORKTREE_PATH_ARGUMENT_INDEX]
+        validate_record(record, schema_path)
+    except ProtocolValidationError as error:
+        print(f"protocol validation failed: {error}", file=sys.stderr)
+        return EXIT_CODE_INVALID_RECORD
+    print(PROTOCOL_VALIDATION_PASSED)
+    return EXIT_CODE_VALID_RECORD
+
+
+if __name__ == "__main__":
+    raise SystemExit(main(sys.argv))