@aws/ml-container-creator 1.0.3 → 1.1.0

package/src/agent/config_loader.py

@@ -0,0 +1,215 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Agent configuration loader.
+
+Reads config/agent.json, applies MCC_* environment variable overrides,
+validates values, and returns a resolved AgentConfig dataclass.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Callable
+
+
+@dataclass(frozen=True)
+class AgentConfig:
+    """Resolved agent configuration (immutable after creation)."""
+
+    model_id: str
+    mcp_servers: list[str]
+    input_cost_per_1k: float
+    output_cost_per_1k: float
+    exit_commands: list[str]
+    reload_commands: list[str]
+    mcp_server_timeout: int
+
+
+_DEFAULTS = AgentConfig(
+    model_id="us.anthropic.claude-sonnet-4-20250514",
+    mcp_servers=[
+        "instance-sizer",
+        "base-image-picker",
+        "model-picker",
+        "workload-picker",
+        "e2e-status",
+        "agent-knowledge",
+    ],
+    input_cost_per_1k=0.003,
+    output_cost_per_1k=0.015,
+    exit_commands=["exit", "quit", "bye", "q"],
+    reload_commands=["reload"],
+    mcp_server_timeout=30,
+)
+
+
+def _warn(msg: str) -> None:
+    """Emit a config warning to stderr."""
+    print(f"[config] warning: {msg}", file=sys.stderr)
+
+
+def _resolve_field(
+    field_name: str,
+    env_var: str | None,
+    file_value: Any | None,
+    default_value: Any,
+    parser: Callable[[str], Any],
+    validator: Callable[[Any], bool],
+) -> Any:
+    """Resolve a single config field using the precedence chain.
+
+    1. Try env var → parse → validate
+    2. Try file value → validate
+    3. Return default
+    """
+    # 1. Environment override
+    if env_var:
+        raw = os.environ.get(env_var)
+        if raw is not None:
+            try:
+                parsed = parser(raw)
+                if validator(parsed):
+                    return parsed
+                else:
+                    _warn(f"{env_var}={raw!r} failed validation, skipping")
+            except (ValueError, TypeError) as e:
+                _warn(f"{env_var}={raw!r} cannot be parsed: {e}")
+
+    # 2. Config file value
+    if file_value is not None:
+        if validator(file_value):
+            return file_value
+        else:
+            _warn(
+                f"config field '{field_name}' has invalid value {file_value!r}, using default"
+            )
+
+    # 3. Hardcoded default
+    return default_value
+
+
+def load_agent_config(config_path: Path | None = None) -> AgentConfig:
+    """Load, validate, and resolve agent configuration.
+
+    Resolution order per parameter (highest to lowest):
+      1. MCC_* environment variable (if set and valid)
+      2. Value from config/agent.json (if file exists and value is valid)
+      3. Hardcoded default
+
+    Args:
+        config_path: Override path to the JSON config file.
+                     Defaults to <package_root>/config/agent.json.
+
+    Returns:
+        Fully-resolved AgentConfig instance.
+    """
+    if config_path is None:
+        package_root = Path(__file__).resolve().parent.parent.parent
+        config_path = package_root / "config" / "agent.json"
+
+    # Read and parse config file
+    file_data: dict[str, Any] = {}
+    if config_path.exists():
+        try:
+            file_data = json.loads(config_path.read_text(encoding="utf-8"))
+        except (json.JSONDecodeError, ValueError) as e:
+            _warn(f"config file '{config_path}' contains invalid JSON: {e}")
+    else:
+        _warn(f"config file '{config_path}' not found, using defaults")
+
+    # Parsers
+    _parse_str: Callable[[str], str] = lambda x: x
+    _parse_list: Callable[[str], list[str]] = lambda raw: [
+        s.strip() for s in raw.split(",")
+    ]
+    _parse_float: Callable[[str], float] = float
+    _parse_int: Callable[[str], int] = int
+
+    # Validators
+    _valid_str: Callable[[Any], bool] = lambda v: isinstance(v, str)
+    _valid_list_str: Callable[[Any], bool] = lambda v: isinstance(v, list) and all(
+        isinstance(s, str) for s in v
+    )
+    _valid_float_nn: Callable[[Any], bool] = (
+        lambda v: isinstance(v, (int, float)) and v >= 0
+    )
+    _valid_int_nn: Callable[[Any], bool] = lambda v: isinstance(v, int) and v >= 0
+
+    # Resolve each field
+    model_id = _resolve_field(
+        "modelId",
+        "MCC_MODEL_ID",
+        file_data.get("modelId"),
+        _DEFAULTS.model_id,
+        _parse_str,
+        _valid_str,
+    )
+
+    mcp_servers = _resolve_field(
+        "mcpServers",
+        "MCC_MCP_SERVERS",
+        file_data.get("mcpServers"),
+        _DEFAULTS.mcp_servers,
+        _parse_list,
+        _valid_list_str,
+    )
+
+    input_cost_per_1k = _resolve_field(
+        "inputCostPer1k",
+        "MCC_INPUT_COST_PER_1K",
+        file_data.get("inputCostPer1k"),
+        _DEFAULTS.input_cost_per_1k,
+        _parse_float,
+        _valid_float_nn,
+    )
+
+    output_cost_per_1k = _resolve_field(
+        "outputCostPer1k",
+        "MCC_OUTPUT_COST_PER_1K",
+        file_data.get("outputCostPer1k"),
+        _DEFAULTS.output_cost_per_1k,
+        _parse_float,
+        _valid_float_nn,
+    )
+
+    exit_commands = _resolve_field(
+        "exitCommands",
+        "MCC_EXIT_COMMANDS",
+        file_data.get("exitCommands"),
+        _DEFAULTS.exit_commands,
+        _parse_list,
+        _valid_list_str,
+    )
+
+    reload_commands = _resolve_field(
+        "reloadCommands",
+        "MCC_RELOAD_COMMANDS",
+        file_data.get("reloadCommands"),
+        _DEFAULTS.reload_commands,
+        _parse_list,
+        _valid_list_str,
+    )
+
+    mcp_server_timeout = _resolve_field(
+        "mcpServerTimeout",
+        "MCC_MCP_SERVER_TIMEOUT",
+        file_data.get("mcpServerTimeout"),
+        _DEFAULTS.mcp_server_timeout,
+        _parse_int,
+        _valid_int_nn,
+    )
+
+    return AgentConfig(
+        model_id=model_id,
+        mcp_servers=mcp_servers,
+        input_cost_per_1k=input_cost_per_1k,
+        output_cost_per_1k=output_cost_per_1k,
+        exit_commands=exit_commands,
+        reload_commands=reload_commands,
+        mcp_server_timeout=mcp_server_timeout,
+    )

package/src/agent/context.py

@@ -0,0 +1,380 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Project context reader for the Strands agent.
+
+Parses all project configuration files into a structured dict for prompt injection.
+Pure file I/O and regex/YAML parsing — no subprocess calls.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+from pathlib import Path
+from typing import Any
+
+import yaml
+
+
+# Regex patterns for shell export parsing
+# Matches: export KEY="VALUE"  or  export KEY='VALUE'  or  export KEY=VALUE
+_EXPORT_QUOTED_RE = re.compile(
+    r"""^export\s+([A-Za-z_][A-Za-z0-9_]*)=["'](.*)["']\s*$"""
+)
+# Matches: export KEY=${KEY:-DEFAULT}
+_EXPORT_DEFAULT_RE = re.compile(
+    r"""^export\s+([A-Za-z_][A-Za-z0-9_]*)=\$\{[A-Za-z_][A-Za-z0-9_]*:-([^}]*)\}\s*$"""
+)
+# Matches: export KEY=VALUE (unquoted, no spaces in value)
+_EXPORT_BARE_RE = re.compile(
+    r"""^export\s+([A-Za-z_][A-Za-z0-9_]*)=([^\s"'$][^\s]*)\s*$"""
+)
+# Matches: export KEY="" (empty quoted value)
+_EXPORT_EMPTY_RE = re.compile(
+    r"""^export\s+([A-Za-z_][A-Za-z0-9_]*)=["']["']\s*$"""
+)
+
+# Dockerfile patterns
+_FROM_RE = re.compile(r"^FROM\s+(.+?)(?:\s+AS\s+\S+)?\s*$", re.IGNORECASE)
+_ENTRYPOINT_RE = re.compile(r"^ENTRYPOINT\s+(.+)\s*$", re.IGNORECASE)
+
+
+class ProjectContext:
+    """Reads and structures all project configuration for the agent.
+
+    Parses do/config, do/ic/*.conf, do/training/config.yaml, Dockerfile,
+    do/adapters/*.conf, the bootstrap profile, and user context files into
+    a unified dict suitable for LLM prompt injection.
+    """
+
+    def __init__(self, project_dir: str) -> None:
+        """Initialize with the project root directory.
+
+        Args:
+            project_dir: Absolute or relative path to the project root
+                         (the directory containing do/config).
+        """
+        self.project_dir = Path(project_dir).resolve()
+
+    def load(self) -> dict[str, Any]:
+        """Load all context. Returns structured dict for prompt injection.
+
+        Gracefully handles missing files — partial context is returned with
+        a ``_missing`` field listing files that could not be parsed.
+
+        Returns:
+            Dict with project configuration structured for prompt injection.
+        """
+        missing: list[str] = []
+
+        do_config = self._parse_do_config(missing)
+        ic_env_vars = self._parse_ic_confs(missing)
+        training_config = self._parse_training_config(missing)
+        dockerfile_info = self._parse_dockerfile(missing)
+        adapters = self._parse_adapters(missing)
+        profile = self._load_profile(missing)
+        user_context = self._load_user_context(missing)
+
+        context: dict[str, Any] = {
+            "project_name": do_config.get("PROJECT_NAME"),
+            "engine": do_config.get("MODEL_SERVER"),
+            "deployment_target": do_config.get("DEPLOYMENT_TARGET"),
+            "model": do_config.get("HF_MODEL_ID") or do_config.get("MODEL_NAME"),
+            "instance_type": do_config.get("INSTANCE_TYPE"),
+            "aws_region": do_config.get("AWS_REGION"),
+            "lora_enabled": do_config.get("ENABLE_LORA", "").lower() == "true",
+            "existing_endpoint": do_config.get("ENDPOINT_NAME")
+            if do_config.get("ENDPOINT_EXTERNAL") == "true"
+            else None,
+            "do_config_vars": do_config,
+            "ic_env_vars": ic_env_vars,
+            "training_config": training_config,
+            "base_image": dockerfile_info.get("base_image"),
+            "entrypoint": dockerfile_info.get("entrypoint"),
+            "adapters": adapters,
+            "profile": profile,
+            "user_context": user_context,
+        }
+
+        if missing:
+            context["_missing"] = missing
+
+        return context
+
+    def _parse_do_config(self, missing: list[str]) -> dict[str, str]:
+        """Parse do/config — regex for export KEY=VALUE lines.
+
+        Handles:
+          - export KEY="VALUE"
+          - export KEY='VALUE'
+          - export KEY=${KEY:-DEFAULT}
+          - export KEY=VALUE (bare, no spaces)
+          - Multi-line values via single-quoted heredoc-style (rare but possible)
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            Dict of variable names to their values.
+        """
+        config_path = self.project_dir / "do" / "config"
+        if not config_path.is_file():
+            missing.append("do/config")
+            return {}
+
+        return self._parse_shell_exports(config_path)
+
+    def _parse_ic_confs(self, missing: list[str]) -> dict[str, dict[str, str]]:
+        """Parse do/ic/*.conf — IC_ENV_* variables grouped by filename.
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            Dict mapping conf filename (without .conf) to a dict of variables.
+        """
+        ic_dir = self.project_dir / "do" / "ic"
+        if not ic_dir.is_dir():
+            missing.append("do/ic/")
+            return {}
+
+        result: dict[str, dict[str, str]] = {}
+        conf_files = sorted(ic_dir.glob("*.conf"))
+
+        if not conf_files:
+            missing.append("do/ic/*.conf")
+            return {}
+
+        for conf_path in conf_files:
+            name = conf_path.stem
+            result[name] = self._parse_shell_exports(conf_path)
+
+        return result
+
+    def _parse_training_config(self, missing: list[str]) -> dict[str, Any] | None:
+        """Parse do/training/config.yaml via yaml.safe_load().
+
+        Extracts key fields: technique, instance_type, hyperparameters,
+        dataset, image, and any other top-level keys.
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            Parsed YAML dict, or None if file is missing/invalid.
+        """
+        yaml_path = self.project_dir / "do" / "training" / "config.yaml"
+        if not yaml_path.is_file():
+            missing.append("do/training/config.yaml")
+            return None
+
+        try:
+            text = yaml_path.read_text(encoding="utf-8")
+            data = yaml.safe_load(text)
+            if not isinstance(data, dict):
+                missing.append("do/training/config.yaml (invalid format)")
+                return None
+            return data
+        except (yaml.YAMLError, OSError):
+            missing.append("do/training/config.yaml (parse error)")
+            return None
+
+    def _parse_dockerfile(self, missing: list[str]) -> dict[str, str | None]:
+        """Extract FROM image and ENTRYPOINT from Dockerfile.
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            Dict with 'base_image' and 'entrypoint' keys.
+        """
+        dockerfile_path = self.project_dir / "Dockerfile"
+        if not dockerfile_path.is_file():
+            missing.append("Dockerfile")
+            return {"base_image": None, "entrypoint": None}
+
+        try:
+            lines = dockerfile_path.read_text(encoding="utf-8").splitlines()
+        except OSError:
+            missing.append("Dockerfile (read error)")
+            return {"base_image": None, "entrypoint": None}
+
+        base_image: str | None = None
+        entrypoint: str | None = None
+
+        for line in lines:
+            stripped = line.strip()
+
+            # Take the last FROM (multi-stage build — final stage is what runs)
+            match = _FROM_RE.match(stripped)
+            if match:
+                base_image = match.group(1).strip()
+
+            match = _ENTRYPOINT_RE.match(stripped)
+            if match:
+                entrypoint = match.group(1).strip()
+
+        return {"base_image": base_image, "entrypoint": entrypoint}
+
+    def _parse_adapters(self, missing: list[str]) -> list[dict[str, Any]]:
+        """List do/adapters/*.conf with adapter names and key variables.
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            List of dicts with 'name' and 'vars' for each adapter conf file.
+        """
+        adapters_dir = self.project_dir / "do" / "adapters"
+        if not adapters_dir.is_dir():
+            missing.append("do/adapters/")
+            return []
+
+        conf_files = sorted(adapters_dir.glob("*.conf"))
+        if not conf_files:
+            missing.append("do/adapters/*.conf")
+            return []
+
+        adapters: list[dict[str, Any]] = []
+        for conf_path in conf_files:
+            name = conf_path.stem
+            variables = self._parse_shell_exports(conf_path)
+            adapters.append({"name": name, "vars": variables})
+
+        return adapters
+
+    def _load_profile(self, missing: list[str]) -> dict[str, Any] | None:
+        """Load bootstrap profile from ~/.ml-container-creator/config.json.
+
+        Reads the config file, finds the active profile, and returns its
+        config object.
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            Active profile config dict, or None if unavailable.
+        """
+        config_path = Path.home() / ".ml-container-creator" / "config.json"
+        if not config_path.is_file():
+            missing.append("~/.ml-container-creator/config.json")
+            return None
+
+        try:
+            text = config_path.read_text(encoding="utf-8")
+            data = json.loads(text)
+        except (json.JSONDecodeError, OSError):
+            missing.append("~/.ml-container-creator/config.json (parse error)")
+            return None
+
+        if not isinstance(data, dict):
+            missing.append("~/.ml-container-creator/config.json (invalid format)")
+            return None
+
+        active_name = data.get("activeProfile")
+        profiles = data.get("profiles")
+
+        if not active_name or not isinstance(profiles, dict):
+            return {"_raw": data, "_note": "no active profile set"}
+
+        profile_config = profiles.get(active_name)
+        if profile_config is None:
+            return {"_raw": data, "_note": f"active profile '{active_name}' not found in profiles"}
+
+        return {"name": active_name, "config": profile_config}
+
+    def _load_user_context(self, missing: list[str]) -> str | None:
+        """Read .mlcc-agent-context.md if it exists in project root.
+
+        This file allows teams to inject custom patterns, conventions,
+        and project-specific guidance into the agent's system prompt.
+
+        Args:
+            missing: Accumulator list for files that could not be found/parsed.
+
+        Returns:
+            File contents as string, or None if file doesn't exist.
+        """
+        context_path = self.project_dir / ".mlcc-agent-context.md"
+        if not context_path.is_file():
+            # This is optional — do not add to missing
+            return None
+
+        try:
+            return context_path.read_text(encoding="utf-8")
+        except OSError:
+            missing.append(".mlcc-agent-context.md (read error)")
+            return None
+
+    def _parse_shell_exports(self, file_path: Path) -> dict[str, str]:
+        """Parse shell export statements from a file.
+
+        Handles multiple patterns:
+          - export KEY="VALUE"
+          - export KEY='VALUE'
+          - export KEY=${KEY:-DEFAULT}
+          - export KEY=BARE_VALUE
+          - Multi-line values with trailing backslash continuation
+
+        Lines starting with # are treated as comments and skipped.
+        Lines that are not export statements are skipped.
+
+        Args:
+            file_path: Path to the shell file to parse.
+
+        Returns:
+            Dict of variable names to their string values.
+        """
+        try:
+            content = file_path.read_text(encoding="utf-8")
+        except OSError:
+            return {}
+
+        variables: dict[str, str] = {}
+        lines = content.splitlines()
+        i = 0
+
+        while i < len(lines):
+            line = lines[i]
+            stripped = line.strip()
+
+            # Skip comments and empty lines
+            if not stripped or stripped.startswith("#"):
+                i += 1
+                continue
+
+            # Handle line continuation (trailing backslash)
+            while stripped.endswith("\\") and i + 1 < len(lines):
+                i += 1
+                stripped = stripped[:-1] + lines[i].strip()
+
+            # Try each pattern in order of specificity
+            match = _EXPORT_EMPTY_RE.match(stripped)
+            if match:
+                variables[match.group(1)] = ""
+                i += 1
+                continue
+
+            match = _EXPORT_QUOTED_RE.match(stripped)
+            if match:
+                variables[match.group(1)] = match.group(2)
+                i += 1
+                continue
+
+            match = _EXPORT_DEFAULT_RE.match(stripped)
+            if match:
+                variables[match.group(1)] = match.group(2)
+                i += 1
+                continue
+
+            match = _EXPORT_BARE_RE.match(stripped)
+            if match:
+                variables[match.group(1)] = match.group(2)
+                i += 1
+                continue
+
+            i += 1
+
+        return variables