Coverage for little_loops / fsm / schema.py: 36%
248 statements
« prev ^ index » next coverage.py v7.12.0, created at 2026-03-18 16:20 -0500
« prev ^ index » next coverage.py v7.12.0, created at 2026-03-18 16:20 -0500
1"""FSM loop schema dataclasses.
3This module defines the type-safe dataclasses that represent FSM loop
4definitions. These match the universal FSM schema described in the
5design documentation.
7The schema supports:
8- Multiple evaluator types (exit_code, output_numeric, etc.)
9- Two-layer transition system (evaluate + route)
10- Both shorthand (on_success/on_failure) and full routing syntax
11- Context variables and captured values
12- LLM evaluation configuration
13"""
15from __future__ import annotations
17from dataclasses import dataclass, field
18from typing import Any, Literal
20# Default LLM model for structured evaluation
21DEFAULT_LLM_MODEL: str = "sonnet"
24@dataclass
25class EvaluateConfig:
26 """Evaluator configuration for action result interpretation.
28 The evaluator determines how to interpret an action's output and
29 produce a verdict string for routing.
31 Attributes:
32 type: Evaluator type. One of:
33 - exit_code: Map exit codes to verdicts (default for shell)
34 - output_numeric: Compare numeric output to target
35 - output_json: Extract and compare JSON path value
36 - output_contains: Pattern matching on stdout
37 - convergence: Compare current vs previous value toward target
38 - llm_structured: Use LLM with structured output (default for slash)
39 operator: Comparison operator (eq, ne, lt, le, gt, ge)
40 target: Target value for comparison
41 tolerance: Acceptable distance from target (for convergence)
42 pattern: Pattern string for output_contains
43 negate: If True, invert the match result (output_contains)
44 path: JSON path for output_json (jq-style)
45 prompt: Custom prompt for llm_structured
46 schema: Custom JSON schema for llm_structured response
47 min_confidence: Minimum confidence threshold for llm_structured
48 uncertain_suffix: If True, append _uncertain to low-confidence verdicts
49 source: Override default source (current action output)
50 previous: Previous value reference for convergence
51 direction: Optimization direction for convergence (minimize/maximize)
52 scope: Paths to limit git diff to for diff_stall evaluator
53 max_stall: Consecutive no-change iterations before failure (diff_stall)
54 """
56 type: Literal[
57 "exit_code",
58 "output_numeric",
59 "output_json",
60 "output_contains",
61 "convergence",
62 "diff_stall",
63 "llm_structured",
64 "mcp_result",
65 ]
66 operator: str | None = None
67 target: int | float | str | None = None
68 tolerance: float | str | None = None # str for interpolation (e.g., "${context.tolerance}")
69 pattern: str | None = None
70 negate: bool = False
71 path: str | None = None
72 prompt: str | None = None
73 schema: dict[str, Any] | None = None
74 min_confidence: float = 0.5
75 uncertain_suffix: bool = False
76 source: str | None = None
77 previous: str | None = None
78 direction: Literal["minimize", "maximize"] = "minimize"
79 scope: list[str] | None = None # for diff_stall: limit git diff to these paths
80 max_stall: int = 1 # for diff_stall: consecutive no-change iterations before failure
82 def to_dict(self) -> dict[str, Any]:
83 """Convert to dictionary for JSON/YAML serialization."""
84 result: dict[str, Any] = {"type": self.type}
86 # Only include non-None optional fields
87 if self.operator is not None:
88 result["operator"] = self.operator
89 if self.target is not None:
90 result["target"] = self.target
91 if self.tolerance is not None:
92 result["tolerance"] = self.tolerance
93 if self.pattern is not None:
94 result["pattern"] = self.pattern
95 if self.negate:
96 result["negate"] = self.negate
97 if self.path is not None:
98 result["path"] = self.path
99 if self.prompt is not None:
100 result["prompt"] = self.prompt
101 if self.schema is not None:
102 result["schema"] = self.schema
103 if self.min_confidence != 0.5:
104 result["min_confidence"] = self.min_confidence
105 if self.uncertain_suffix:
106 result["uncertain_suffix"] = self.uncertain_suffix
107 if self.source is not None:
108 result["source"] = self.source
109 if self.previous is not None:
110 result["previous"] = self.previous
111 if self.direction != "minimize":
112 result["direction"] = self.direction
113 if self.scope is not None:
114 result["scope"] = self.scope
115 if self.max_stall != 1:
116 result["max_stall"] = self.max_stall
118 return result
120 @classmethod
121 def from_dict(cls, data: dict[str, Any]) -> EvaluateConfig:
122 """Create from dictionary (JSON/YAML deserialization)."""
123 return cls(
124 type=data["type"],
125 operator=data.get("operator"),
126 target=data.get("target"),
127 tolerance=data.get("tolerance"),
128 pattern=data.get("pattern"),
129 negate=data.get("negate", False),
130 path=data.get("path"),
131 prompt=data.get("prompt"),
132 schema=data.get("schema"),
133 min_confidence=data.get("min_confidence", 0.5),
134 uncertain_suffix=data.get("uncertain_suffix", False),
135 source=data.get("source"),
136 previous=data.get("previous"),
137 direction=data.get("direction", "minimize"),
138 scope=data.get("scope"),
139 max_stall=data.get("max_stall", 1),
140 )
143@dataclass
144class RouteConfig:
145 """Routing table configuration for verdict-to-state mapping.
147 Maps verdict strings from evaluators to next state names.
149 Attributes:
150 routes: Mapping from verdict string to next state name
151 default: Default state for unmatched verdicts (the "_" key)
152 error: State for evaluation/execution errors (the "_error" key)
153 """
155 routes: dict[str, str] = field(default_factory=dict)
156 default: str | None = None
157 error: str | None = None
159 def to_dict(self) -> dict[str, Any]:
160 """Convert to dictionary for JSON/YAML serialization."""
161 result = dict(self.routes)
162 if self.default is not None:
163 result["_"] = self.default
164 if self.error is not None:
165 result["_error"] = self.error
166 return result
168 @classmethod
169 def from_dict(cls, data: dict[str, Any]) -> RouteConfig:
170 """Create from dictionary (JSON/YAML deserialization)."""
171 routes = {k: v for k, v in data.items() if not k.startswith("_")}
172 return cls(
173 routes=routes,
174 default=data.get("_"),
175 error=data.get("_error"),
176 )
179@dataclass
180class StateConfig:
181 """Configuration for a single FSM state.
183 States can have actions, evaluators, and routing. Supports both
184 shorthand (on_success/on_failure) and full routing table syntax.
186 Attributes:
187 action: Command to execute (shell, slash command, or "server/tool-name" for mcp_tool)
188 action_type: How to execute the action (prompt, slash_command, shell, mcp_tool).
189 If None, uses heuristic: / prefix = slash_command, else = shell.
190 params: MCP tool arguments (only used with action_type: mcp_tool). Supports
191 ${variable} interpolation in string values.
192 evaluate: Evaluator configuration for result interpretation
193 route: Full routing table (verdict -> state mapping)
194 on_yes: Shorthand for yes verdict routing
195 on_no: Shorthand for no verdict routing
196 on_error: Shorthand for error verdict routing
197 on_partial: Shorthand for partial verdict routing
198 next: Unconditional transition (no evaluation)
199 terminal: If True, this is an end state
200 capture: Variable name to store action output
201 timeout: Action-level timeout in seconds
202 on_maintain: State to transition to when maintain=True and loop completes
203 max_retries: Max consecutive re-entries before transitioning to on_retry_exhausted.
204 A value of N allows N retries after the initial execution (N+1 total entries).
205 Requires on_retry_exhausted to also be set.
206 on_retry_exhausted: State to transition to when max_retries consecutive re-entries
207 are exceeded. Required when max_retries is set.
208 loop: Name of a loop YAML to execute as a sub-FSM. Mutually exclusive with action.
209 context_passthrough: When True, pass parent context variables to child loop and
210 merge child captures back into parent context.
211 """
213 action: str | None = None
214 action_type: Literal["prompt", "slash_command", "shell", "mcp_tool"] | None = None
215 params: dict[str, Any] = field(default_factory=dict)
216 evaluate: EvaluateConfig | None = None
217 route: RouteConfig | None = None
218 on_yes: str | None = None
219 on_no: str | None = None
220 on_error: str | None = None
221 on_partial: str | None = None
222 on_blocked: str | None = None
223 next: str | None = None
224 terminal: bool = False
225 capture: str | None = None
226 timeout: int | None = None
227 on_maintain: str | None = None
228 max_retries: int | None = None
229 on_retry_exhausted: str | None = None
230 loop: str | None = None
231 context_passthrough: bool = False
233 def to_dict(self) -> dict[str, Any]:
234 """Convert to dictionary for JSON/YAML serialization."""
235 result: dict[str, Any] = {}
237 if self.action is not None:
238 result["action"] = self.action
239 if self.action_type is not None:
240 result["action_type"] = self.action_type
241 if self.params:
242 result["params"] = self.params
243 if self.evaluate is not None:
244 result["evaluate"] = self.evaluate.to_dict()
245 if self.route is not None:
246 result["route"] = self.route.to_dict()
247 if self.on_yes is not None:
248 result["on_yes"] = self.on_yes
249 if self.on_no is not None:
250 result["on_no"] = self.on_no
251 if self.on_error is not None:
252 result["on_error"] = self.on_error
253 if self.on_partial is not None:
254 result["on_partial"] = self.on_partial
255 if self.on_blocked is not None:
256 result["on_blocked"] = self.on_blocked
257 if self.next is not None:
258 result["next"] = self.next
259 if self.terminal:
260 result["terminal"] = self.terminal
261 if self.capture is not None:
262 result["capture"] = self.capture
263 if self.timeout is not None:
264 result["timeout"] = self.timeout
265 if self.on_maintain is not None:
266 result["on_maintain"] = self.on_maintain
267 if self.max_retries is not None:
268 result["max_retries"] = self.max_retries
269 if self.on_retry_exhausted is not None:
270 result["on_retry_exhausted"] = self.on_retry_exhausted
271 if self.loop is not None:
272 result["loop"] = self.loop
273 if self.context_passthrough:
274 result["context_passthrough"] = self.context_passthrough
276 return result
278 @classmethod
279 def from_dict(cls, data: dict[str, Any]) -> StateConfig:
280 """Create from dictionary (JSON/YAML deserialization)."""
281 evaluate = None
282 if "evaluate" in data:
283 evaluate = EvaluateConfig.from_dict(data["evaluate"])
285 route = None
286 if "route" in data:
287 route = RouteConfig.from_dict(data["route"])
289 return cls(
290 action=data.get("action"),
291 action_type=data.get("action_type"),
292 params=data.get("params", {}),
293 evaluate=evaluate,
294 route=route,
295 on_yes=data.get("on_yes") or data.get("on_success"),
296 on_no=data.get("on_no") or data.get("on_failure"),
297 on_error=data.get("on_error"),
298 on_partial=data.get("on_partial"),
299 on_blocked=data.get("on_blocked"),
300 next=data.get("next"),
301 terminal=data.get("terminal", False),
302 capture=data.get("capture"),
303 timeout=data.get("timeout"),
304 on_maintain=data.get("on_maintain"),
305 max_retries=data.get("max_retries"),
306 on_retry_exhausted=data.get("on_retry_exhausted"),
307 loop=data.get("loop"),
308 context_passthrough=data.get("context_passthrough", False),
309 )
311 def get_referenced_states(self) -> set[str]:
312 """Get all state names referenced by this state configuration.
314 Returns:
315 Set of state names that this state can transition to.
316 """
317 refs: set[str] = set()
319 if self.on_yes is not None:
320 refs.add(self.on_yes)
321 if self.on_no is not None:
322 refs.add(self.on_no)
323 if self.on_error is not None:
324 refs.add(self.on_error)
325 if self.on_partial is not None:
326 refs.add(self.on_partial)
327 if self.on_blocked is not None:
328 refs.add(self.on_blocked)
329 if self.next is not None:
330 refs.add(self.next)
331 if self.on_maintain is not None:
332 refs.add(self.on_maintain)
333 if self.on_retry_exhausted is not None:
334 refs.add(self.on_retry_exhausted)
335 if self.route is not None:
336 refs.update(self.route.routes.values())
337 if self.route.default is not None:
338 refs.add(self.route.default)
339 if self.route.error is not None:
340 refs.add(self.route.error)
342 return refs
345@dataclass
346class LLMConfig:
347 """LLM evaluation configuration.
349 Settings for the llm_structured evaluator.
351 Attributes:
352 enabled: If False, disable LLM evaluation entirely
353 model: Model identifier for LLM calls
354 max_tokens: Maximum tokens for evaluation response
355 timeout: Timeout for LLM calls in seconds
356 """
358 enabled: bool = True
359 model: str = DEFAULT_LLM_MODEL
360 max_tokens: int = 256
361 timeout: int = 1800
363 def to_dict(self) -> dict[str, Any]:
364 """Convert to dictionary for JSON/YAML serialization."""
365 result: dict[str, Any] = {}
367 if not self.enabled:
368 result["enabled"] = self.enabled
369 if self.model != DEFAULT_LLM_MODEL:
370 result["model"] = self.model
371 if self.max_tokens != 256:
372 result["max_tokens"] = self.max_tokens
373 if self.timeout != 1800:
374 result["timeout"] = self.timeout
376 return result if result else {}
378 @classmethod
379 def from_dict(cls, data: dict[str, Any]) -> LLMConfig:
380 """Create from dictionary (JSON/YAML deserialization)."""
381 return cls(
382 enabled=data.get("enabled", True),
383 model=data.get("model", DEFAULT_LLM_MODEL),
384 max_tokens=data.get("max_tokens", 256),
385 timeout=data.get("timeout", 1800),
386 )
389@dataclass
390class FSMLoop:
391 """Complete FSM loop definition.
393 The main dataclass representing a loop configuration.
395 Attributes:
396 name: Unique loop identifier
397 initial: Starting state name
398 states: Mapping from state name to StateConfig
399 context: User-defined shared variables
400 scope: Paths this loop operates on (for concurrency control)
401 max_iterations: Safety limit for loop iterations
402 backoff: Seconds between iterations
403 timeout: Max total runtime in seconds (loop-level)
404 maintain: If True, restart after completion
405 llm: LLM evaluation configuration
406 on_handoff: Behavior when handoff signal detected (pause/spawn/terminate)
407 """
409 name: str
410 initial: str
411 states: dict[str, StateConfig]
412 description: str | None = None
413 context: dict[str, Any] = field(default_factory=dict)
414 scope: list[str] = field(default_factory=list)
415 max_iterations: int = 50
416 backoff: float | None = None
417 timeout: int | None = None
418 default_timeout: int | None = None
419 maintain: bool = False
420 llm: LLMConfig = field(default_factory=LLMConfig)
421 on_handoff: Literal["pause", "spawn", "terminate"] = "pause"
422 input_key: str = "input"
424 def to_dict(self) -> dict[str, Any]:
425 """Convert to dictionary for JSON/YAML serialization."""
426 result: dict[str, Any] = {
427 "name": self.name,
428 "initial": self.initial,
429 "states": {name: state.to_dict() for name, state in self.states.items()},
430 }
432 if self.description is not None:
433 result["description"] = self.description
434 if self.context:
435 result["context"] = self.context
436 if self.scope:
437 result["scope"] = self.scope
438 if self.max_iterations != 50:
439 result["max_iterations"] = self.max_iterations
440 if self.backoff is not None:
441 result["backoff"] = self.backoff
442 if self.timeout is not None:
443 result["timeout"] = self.timeout
444 if self.default_timeout is not None:
445 result["default_timeout"] = self.default_timeout
446 if self.maintain:
447 result["maintain"] = self.maintain
448 if self.on_handoff != "pause":
449 result["on_handoff"] = self.on_handoff
450 if self.input_key != "input":
451 result["input_key"] = self.input_key
453 llm_dict = self.llm.to_dict()
454 if llm_dict:
455 result["llm"] = llm_dict
457 return result
459 @classmethod
460 def from_dict(cls, data: dict[str, Any]) -> FSMLoop:
461 """Create from dictionary (JSON/YAML deserialization)."""
462 states = {
463 name: StateConfig.from_dict(state_data)
464 for name, state_data in data.get("states", {}).items()
465 }
467 llm = LLMConfig()
468 if "llm" in data:
469 llm = LLMConfig.from_dict(data["llm"])
471 return cls(
472 name=data["name"],
473 initial=data["initial"],
474 states=states,
475 description=data.get("description"),
476 context=data.get("context", {}),
477 scope=data.get("scope", []),
478 max_iterations=data.get("max_iterations", 50),
479 backoff=data.get("backoff"),
480 timeout=data.get("timeout"),
481 default_timeout=data.get("default_timeout"),
482 maintain=data.get("maintain", False),
483 llm=llm,
484 on_handoff=data.get("on_handoff", "pause"),
485 input_key=data.get("input_key", "input"),
486 )
488 def get_all_state_names(self) -> set[str]:
489 """Get all defined state names.
491 Returns:
492 Set of all state names in this FSM.
493 """
494 return set(self.states.keys())
496 def get_terminal_states(self) -> set[str]:
497 """Get all terminal state names.
499 Returns:
500 Set of state names where terminal=True.
501 """
502 return {name for name, state in self.states.items() if state.terminal}
504 def get_all_referenced_states(self) -> set[str]:
505 """Get all state names referenced by transitions.
507 This includes the initial state and all states referenced
508 in routing configurations.
510 Returns:
511 Set of all referenced state names.
512 """
513 refs: set[str] = {self.initial}
514 for state in self.states.values():
515 refs.update(state.get_referenced_states())
516 return refs