agentbyte 0.24.2__tar.gz → 0.26.0__tar.gz

{agentbyte-0.24.2 → agentbyte-0.26.0}/CHANGELOG.md

@@ -4,7 +4,41 @@ All notable changes to Agentbyte are documented in this file.
 
 The format follows Keep a Changelog principles and semantic versioning.
 
-## [Unreleased]
+## [0.26.0] - 2026-09-04
+
+### Added
+
+- `FunctionTool` now supports structured parameters backed by Pydantic models, including nested models, optional collections, typed dictionaries, tuples, unions, field constraints, and `Annotated` parameter descriptions.
+
+### Changed
+
+- Function-tool schema generation and execution now share one strict Pydantic validation model. Validated object arguments reach Pydantic-annotated functions as model instances, dictionary annotations remain dictionaries, and JSON arrays are converted to declared tuple types.
+- Function tools now reject unexpected arguments and unsupported annotations at the boundary instead of silently accepting or flattening them.
+
+### Fixed
+
+- Fixed optional and model-shaped tool parameters being incorrectly advertised as strings, which caused valid object-shaped LLM tool calls to fail with `Invalid parameters provided` before tool execution.
+- Parameter schemas now use author-provided `Field` or string `Annotated` descriptions instead of leaking Python builtin and typing documentation.
+
+## [0.25.0] - 2026-09-01
+
+### Added
+
+- New `agentbyte.optim` subpackage: an autotune harness for searching an agent's instructions, skills, and tool selection against an `agentbyte.eval` dataset. Includes `AgentConfig`/`OptimizationSpec` with six tunables, `BaseOptimizer`/`CostLog`/`OptimizationTrace`, `ReflectiveOptimizer`, `ReflectiveParetoOptimizer`, `MIPROOptimizer`, and an optional GEPA bridge (`optimize_with_gepa`) behind a new `agentbyte[optim]` extra.
+- Structured skill discovery triggers: `Skill.triggers` (normalized, deduplicated, capped at 20 entries of up to 100 characters each) can now be declared on `InlineSkill`, `FileSkill` (via `SKILL.md` YAML frontmatter), and `ClassSkill`, and are rendered as nested `<triggers>` elements in the provider's skill advertisement.
+- `AgentSkillConfig` in `agentbyte.optim` gained a `triggers` field and a new `SkillTriggersTunable`, so trigger phrases can be searched alongside instructions and descriptions.
+
+### Changed
+
+- **Breaking:** `AgentConfig.skills` now requires structured `AgentSkillConfig` values (`description` + `instructions`, both non-empty) instead of plain instruction strings; legacy string values are rejected.
+- Skill providers now register only the tools (`load_skill`, `read_skill_resource`, `run_skill_script`) actually needed by their resolved skills, instead of always registering all three.
+- Duplicate skill/resource/script names across providers and file roots now resolve deterministically via case-insensitive, first-declared-wins precedence.
+
+### Fixed
+
+- `AgentConfig.model` is now honored by `to_agent()` — previously the field was accepted but silently ignored, always reusing the passed-in `model_client` unchanged. The override now round-trips through the client's `_to_config()`/`_from_config()` methods.
+- `FileSkillsSource` skill-directory discovery now sorts directory entries before recursing, making duplicate-name resolution deterministic across filesystems/platforms instead of depending on OS-level `iterdir()` ordering.
+- Missing skill/resource/script lookups now return actionable errors listing the available names instead of a bare "not found".
 
 ## [0.24.2] - 2026-08-31
 

{agentbyte-0.24.2 → agentbyte-0.26.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: agentbyte
-Version: 0.24.2
+Version: 0.26.0
 Summary: A toolkit for designing multiagent systems
 Author-email: MrDataPsycho <mr.data.psycho@gmail.com>
 License-Expression: LicenseRef-Proprietary
@@ -18,6 +18,7 @@ Requires-Dist: asyncpg>=0.31.0; extra == 'all'
 Requires-Dist: azure-identity>=1.25.1; extra == 'all'
 Requires-Dist: beautifulsoup4>=4.12; extra == 'all'
 Requires-Dist: fastapi>=0.135.2; extra == 'all'
+Requires-Dist: gepa>=0.1; extra == 'all'
 Requires-Dist: graphviz>=0.21; extra == 'all'
 Requires-Dist: html2text>=2024.2; extra == 'all'
 Requires-Dist: openai>=1.107.1; extra == 'all'
@@ -46,6 +47,8 @@ Requires-Dist: pymupdf-layout>=1.26.6; extra == 'dev'
 Requires-Dist: pymupdf4llm>=0.2.9; extra == 'dev'
 Provides-Extra: openai
 Requires-Dist: openai>=1.107.1; extra == 'openai'
+Provides-Extra: optim
+Requires-Dist: gepa>=0.1; extra == 'optim'
 Provides-Extra: otel
 Requires-Dist: opentelemetry-api>=1.39.1; extra == 'otel'
 Requires-Dist: opentelemetry-exporter-otlp-proto-http>=1.39.1; extra == 'otel'
@@ -83,7 +86,7 @@ Description-Content-Type: text/markdown
 
 Agentbyte is an observability-first agentic AI framework for building and studying multiagent systems with a learning-first, implementation-oriented workflow.
 
-Current release: **0.24.2**
+Current release: **0.26.0**
 
 ## Building an Agent
 

{agentbyte-0.24.2 → agentbyte-0.26.0}/README.md

@@ -2,7 +2,7 @@
 
 Agentbyte is an observability-first agentic AI framework for building and studying multiagent systems with a learning-first, implementation-oriented workflow.
 
-Current release: **0.24.2**
+Current release: **0.26.0**
 
 ## Building an Agent
 

{agentbyte-0.24.2 → agentbyte-0.26.0}/pyproject.toml

@@ -47,6 +47,9 @@ sql = [
     "asyncpg>=0.31.0",
     "aiosqlite>=0.22.1",
 ]
+optim = [
+    "gepa>=0.1",
+]
 viz = [
     "graphviz>=0.21",
 ]
@@ -97,6 +100,7 @@ all = [
     "html2text>=2024.2",
     "arxiv>=2.1",
     "youtube-transcript-api>=0.6",
+    "gepa>=0.1",
 ]
 
 
@@ -156,6 +160,9 @@ sql = [
     "asyncpg>=0.31.0",
     "aiosqlite>=0.22.1",
 ]
+optim = [
+    "gepa>=0.1",
+]
 test = [
     "pytest>=9.0.1",
     "pytest-asyncio>=1.3.0",
@@ -201,6 +208,7 @@ all = [
     "html2text>=2024.2",
     "arxiv>=2.1",
     "youtube-transcript-api>=0.6",
+    "gepa>=0.1",
 ]
 
 [tool.pytest.ini_options]

agentbyte-0.26.0/src/agentbyte/__about__.py

@@ -0,0 +1,2 @@
+__version__ = "0.26.0"
+VERSION = __version__

{agentbyte-0.24.2 → agentbyte-0.26.0}/src/agentbyte/context_providers/skill_tools.py

@@ -19,6 +19,14 @@ def _find_skill(skills: Sequence[Skill], name: str) -> Skill | None:
     return None
 
 
+def _available_names(kind: str, names: Sequence[str]) -> str:
+    unique_names = {name.lower(): name for name in reversed(names)}
+    available = sorted(unique_names.values(), key=str.lower)
+    if not available:
+        return f"No {kind} are available."
+    return f"Available {kind}: {', '.join(available)}."
+
+
 def _error(message: str) -> ToolResult:
     return ToolResult(success=False, result=message, error=message, metadata={})
 
@@ -57,7 +65,8 @@ class LoadSkillTool(BaseTool):
             return _error("Error: Skill name cannot be empty.")
         skill = _find_skill(self._skills, skill_name)
         if skill is None:
-            return _error(f"Error: Skill '{skill_name}' not found.")
+            available = _available_names("skills", [skill.name for skill in self._skills])
+            return _error(f"Error: Skill '{skill_name}' not found. {available}")
         logger.info("Loading skill: %s", skill_name)
         return _ok(skill.content)
 
@@ -104,7 +113,8 @@ class ReadSkillResourceTool(BaseTool):
 
         skill = _find_skill(self._skills, skill_name)
         if skill is None:
-            return _error(f"Error: Skill '{skill_name}' not found.")
+            available = _available_names("skills", [skill.name for skill in self._skills])
+            return _error(f"Error: Skill '{skill_name}' not found. {available}")
 
         resource_name_lower = resource_name.lower()
         resource = next(
@@ -112,8 +122,12 @@ class ReadSkillResourceTool(BaseTool):
             None,
         )
         if resource is None:
+            available = _available_names(
+                "resources", [resource.name for resource in skill.resources]
+            )
             return _error(
-                f"Error: Resource '{resource_name}' not found in skill '{skill_name}'."
+                f"Error: Resource '{resource_name}' not found in skill '{skill_name}'. "
+                f"{available}"
             )
 
         try:
@@ -183,15 +197,20 @@ class RunSkillScriptTool(BaseTool):
 
         skill = _find_skill(self._skills, skill_name)
         if skill is None:
-            return _error(f"Error: Skill '{skill_name}' not found.")
+            available = _available_names("skills", [skill.name for skill in self._skills])
+            return _error(f"Error: Skill '{skill_name}' not found. {available}")
 
         script = next(
             (s for s in skill.scripts if s.name.lower() == script_name.lower()),
             None,
         )
         if script is None:
+            available = _available_names(
+                "scripts", [script.name for script in skill.scripts]
+            )
             return _error(
-                f"Error: Script '{script_name}' not found in skill '{skill_name}'."
+                f"Error: Script '{script_name}' not found in skill '{skill_name}'. "
+                f"{available}"
             )
 
         try:

{agentbyte-0.24.2 → agentbyte-0.26.0}/src/agentbyte/context_providers/skills.py

@@ -12,7 +12,12 @@ from agentbyte.context_providers.skill_tools import (
 )
 from agentbyte.skills.base import Skill
 from agentbyte.skills.scripts import SkillScriptRunner
-from agentbyte.skills.sources import FileSkillsSource, InMemorySkillsSource, SkillsSource
+from agentbyte.skills.sources import (
+    DeduplicatingSkillsSource,
+    FileSkillsSource,
+    InMemorySkillsSource,
+    SkillsSource,
+)
 from agentbyte.tools import BaseTool
 
 _DEFAULT_TEMPLATE = """\
@@ -47,6 +52,11 @@ def _build_advertisement(skills: Sequence[Skill]) -> str:
         lines.append("  <skill>")
         lines.append(f"    <name>{xml_escape(skill.name)}</name>")
         lines.append(f"    <description>{xml_escape(skill.description)}</description>")
+        if skill.triggers:
+            lines.append("    <triggers>")
+            for trigger in skill.triggers:
+                lines.append(f"      <trigger>{xml_escape(trigger)}</trigger>")
+            lines.append("    </triggers>")
         lines.append("  </skill>")
     return "\n".join(lines)
 
@@ -80,11 +90,12 @@ class SkillsProvider(ContextProvider):
         disable_caching: bool = False,
     ) -> None:
         if isinstance(source, Skill):
-            self._source: SkillsSource = InMemorySkillsSource([source])
+            skill_source: SkillsSource = InMemorySkillsSource([source])
         elif isinstance(source, list):
-            self._source = InMemorySkillsSource(source)
+            skill_source = InMemorySkillsSource(source)
         else:
-            self._source = source
+            skill_source = source
+        self._source = DeduplicatingSkillsSource(skill_source)
         self._instruction_template = instruction_template
         self._require_script_approval = require_script_approval
         self._disable_caching = disable_caching
@@ -132,7 +143,9 @@ class SkillsProvider(ContextProvider):
         tools: list[BaseTool] = []
         if skills:
             tools.append(LoadSkillTool(skills))
+        if has_resources:
             tools.append(ReadSkillResourceTool(skills))
+        if has_scripts:
             tools.append(
                 RunSkillScriptTool(skills, require_approval=self._require_script_approval)
             )

agentbyte-0.26.0/src/agentbyte/optim/__init__.py

@@ -0,0 +1,66 @@
+"""Agent optimization (autotune) for agentbyte.
+
+See docs/specs/0038-agent-optimization-autotune for the full design.
+"""
+
+from .base import BaseOptimizer, Candidate, CostLog, OptimizationResult, task_key
+from .config import AgentConfig, AgentSkillConfig
+from .gepa import GEPA_AVAILABLE, AgentbyteGepaAdapter, GepaRunResult, optimize_with_gepa
+from .mipro import TIPS, MIPROOptimizer
+from .pareto import ReflectiveParetoOptimizer, pareto_frontier
+from .reflective import EditOut, ReflectionOut, ReflectiveOptimizer
+from .spec import (
+    CatalogEntry,
+    Component,
+    Edit,
+    InstructionTunable,
+    Operation,
+    OptimizationSpec,
+    ProposalMode,
+    SkillDescriptionTunable,
+    SkillTunable,
+    SkillTriggersTunable,
+    ToolDescriptionTunable,
+    ToolSelectionTunable,
+    Tunable,
+)
+from .trace import OptimizationTrace, load_trace, serialize_messages, task_record, usage_dict
+
+__all__ = [
+    "AgentConfig",
+    "AgentSkillConfig",
+    "Operation",
+    "ProposalMode",
+    "CatalogEntry",
+    "Component",
+    "Edit",
+    "Tunable",
+    "InstructionTunable",
+    "SkillTunable",
+    "SkillDescriptionTunable",
+    "SkillTriggersTunable",
+    "ToolDescriptionTunable",
+    "ToolSelectionTunable",
+    "OptimizationSpec",
+    "OptimizationTrace",
+    "load_trace",
+    "task_record",
+    "usage_dict",
+    "serialize_messages",
+    "CostLog",
+    "Candidate",
+    "OptimizationResult",
+    "BaseOptimizer",
+    "task_key",
+    "ReflectiveOptimizer",
+    "EditOut",
+    "ReflectionOut",
+    "ReflectiveParetoOptimizer",
+    "pareto_frontier",
+    "MIPROOptimizer",
+    "TIPS",
+    "GEPA_AVAILABLE",
+    "AgentbyteGepaAdapter",
+    "GepaRunResult",
+    "optimize_with_gepa",
+]