File size: 26,408 Bytes
eab1ab3 662e89d d715ed0 eab1ab3 d715ed0 cf32d0c eab1ab3 cf32d0c d715ed0 eab1ab3 d715ed0 cf32d0c d715ed0 cf32d0c eab1ab3 d715ed0 eab1ab3 920fc45 d715ed0 920fc45 662e89d eab1ab3 d715ed0 eab1ab3 6eea076 eab1ab3 6eea076 eab1ab3 6eea076 eab1ab3 d715ed0 08381bd d715ed0 920fc45 d715ed0 eab1ab3 08381bd eab1ab3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 | # ---- Changelog ----
# [2026-04-12] Claude (Sonnet 4.6) β Fix shell_allowlist double-gate bug
# What: spec shell_allowlist was bypassing executor-level policy check but not the
# ShellTool's own internal policy check β command was denied by the second gate.
# Why: use_direct_dispatch was False when no workspace override, routing through
# TOOL_REGISTRY β ShellTool.can_execute_shell() which ignores spec allowlist.
# How: Added third condition to use_direct_dispatch: when _skip_policy_shell is True
# (spec shell_allowlist approved the command), force direct dispatch to bypass
# the tool-level check entirely. Executor already validated; no double-checking.
# [2026-04-06] Josh + Claude β Fix 3 spec executor gaps
# What: (1) Bypass tool-level path checks when workspace differs from default repo path
# (2) Support spec-configurable shell_allowlist in constraints
# (3) Wire edit_file tool through direct dispatch like read/write
# Why: Cross-repo specs fail because FilesystemTool._check_path() enforces repo_path,
# shell allowlist was hardcoded with no spec override, edit_file didn't exist
# How: Direct filesystem ops for read/write/edit when workspace != REPO_PATH,
# shell_allowlist check before PolicyEngine for spec-declared commands
# [2026-04-05] Josh + Claude β Structured spec execution engine
# What: Deterministic execution of WorkBlockSpec JSON specs
# Why: No LLM in the loop β mechanical step execution, mandatory validation, zero improvisation
# How: State machine over step types (action/gate/condition/loop/group), variable bindings,
# uses existing TOOL_REGISTRY and PolicyEngine, produces structured execution reports
# -------------------
"""Spec Executor β mechanical execution of structured work block specs.
The spec IS the plan. The executor does not interpret, reason, or improvise.
It executes steps top-to-bottom, validates after every action, and reports results.
"""
import json
import logging
import re
import time
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any, Optional
from work_block_schema import validate_spec
logger = logging.getLogger("spec_executor")
@dataclass
class ExecutionContext:
"""Tracks state across a block's execution."""
block_id: str
bindings: dict = field(default_factory=dict) # $var -> result string
step_results: dict = field(default_factory=dict) # step_id -> {status, ...}
tool_call_count: int = 0
start_time: float = field(default_factory=time.time)
aborted: bool = False
abort_reason: str = ""
class SpecExecutor:
"""Deterministic executor for WorkBlockSpec JSON specs.
Uses the same TOOL_REGISTRY and PolicyEngine as the conversational agent loop.
No LLM calls β this is a mechanical state machine.
"""
def __init__(self, tool_registry: dict, policy_check_fn, worker_ng, workspace: Path):
"""
Args:
tool_registry: The TOOL_REGISTRY dict from app.py (name -> lambda handler)
policy_check_fn: The check_tool_call function from policy_engine
worker_ng: NeuroGraphMemory instance for substrate ingestion
workspace: Path to the workspace root
"""
self.tools = tool_registry
self.policy_check = policy_check_fn
self.ng = worker_ng
self.workspace = workspace
self._default_workspace = workspace # Original repo path β used to detect workspace overrides
self.read_only_paths: list[Path] = []
def execute_block(self, spec: dict) -> dict:
"""Execute a work block spec. Returns a structured execution report."""
# 1. Schema validation
valid, errors = validate_spec(spec)
if not valid:
return {
"report_version": "1.0.0",
"block_id": spec.get("block", {}).get("id", "UNKNOWN"),
"status": "rejected",
"errors": errors,
}
block = spec["block"]
constraints = spec["constraints"]
ctx = ExecutionContext(block_id=block["id"])
# Read-only external paths from spec constraints (sibling repos QB can read)
self.read_only_paths = [
Path(p).expanduser() for p in constraints.get("read_only_paths", [])
]
if self.read_only_paths:
logger.info("Read-only external paths: %s", self.read_only_paths)
# Use spec-declared workspace if present, otherwise default
if "workspace" in block and block["workspace"]:
self.workspace = Path(block["workspace"]).expanduser().resolve()
logger.info("Workspace override: %s", self.workspace)
logger.info("Executing block %s: %s", block["id"], block["name"])
# 2. Execute steps
for step in spec["steps"]:
if ctx.aborted:
break
if ctx.tool_call_count >= constraints.get("max_iterations", 15):
ctx.aborted = True
ctx.abort_reason = f"Max iterations ({constraints.get('max_iterations', 15)}) exceeded"
break
elapsed = time.time() - ctx.start_time
if elapsed > constraints.get("timeout_seconds", 300):
ctx.aborted = True
ctx.abort_reason = f"Block timeout ({constraints.get('timeout_seconds', 300)}s) exceeded"
break
self._execute_step(step, ctx, constraints)
# 3. Build report
return self._build_report(spec, ctx)
# ------------------------------------------------------------------
# Step dispatch
# ------------------------------------------------------------------
def _execute_step(self, step: dict, ctx: ExecutionContext, constraints: dict):
step_type = step["type"]
dispatch = {
"action": self._execute_action,
"gate": self._execute_gate,
"condition": self._execute_condition,
"loop": self._execute_loop,
"group": self._execute_group,
}
handler = dispatch.get(step_type)
if handler:
handler(step, ctx, constraints)
else:
ctx.step_results[step["id"]] = {"status": "fail", "reason": f"Unknown step type: {step_type}"}
# ------------------------------------------------------------------
# Action steps
# ------------------------------------------------------------------
def _execute_action(self, step: dict, ctx: ExecutionContext, constraints: dict):
tool_name = step["tool"]
step_id = step["id"]
desc = step.get("description", tool_name)
logger.info(" [%s] %s β %s", step_id, desc, tool_name)
# Tool allowlist check
allowlist = constraints.get("tool_allowlist")
if allowlist and tool_name not in allowlist:
self._record_failure(ctx, step_id, f"Tool '{tool_name}' not in block allowlist")
self._handle_failure(step["on_failure"], step_id, ctx)
return
# Resolve $var references in params
params = self._resolve_bindings(step["params"], ctx)
# --- Gap 2: Spec-configurable shell allowlist ---
# If the spec declares a shell_allowlist in constraints and this is a
# shell_execute call, check against the spec's list FIRST. If the command
# matches the spec allowlist, skip PolicyEngine's shell check (the executor
# already ran the broader policy check above for path/content).
_skip_policy_shell = False
if tool_name == "shell_execute":
spec_shell_allowlist = constraints.get("shell_allowlist")
if spec_shell_allowlist:
import os as _os
cmd = (params.get("command") or "").strip()
if cmd:
binary = _os.path.basename(cmd.split()[0])
if any(binary == a or binary.startswith(a) for a in spec_shell_allowlist):
_skip_policy_shell = True
logger.info(" [%s] Shell command '%s' approved by spec allowlist", step_id, binary)
# PolicyEngine rim check β skip shell portion if spec allowlist approved it
if _skip_policy_shell:
# Still run path/content checks, just not the shell check.
# We call policy_check and ignore shell-denied results.
allowed, reason = self.policy_check(tool_name, params, self.workspace, self.read_only_paths)
if not allowed and "not on the allowlist" in reason:
# This is the shell allowlist denial β spec overrides it
allowed, reason = True, "Permitted by spec shell_allowlist."
else:
allowed, reason = self.policy_check(tool_name, params, self.workspace, self.read_only_paths)
if not allowed:
self._record_failure(ctx, step_id, f"PolicyEngine denied: {reason}")
self._handle_failure(step["on_failure"], step_id, ctx)
return
# --- Gap 1: Workspace-aware tool dispatch ---
# When the spec declares a workspace that differs from the default repo path,
# the TOOL_REGISTRY lambdas route through FilesystemTool._check_path() which
# is hardcoded to repo_path. Bypass the registry for filesystem tools and call
# operations directly against the spec's workspace. The executor already did
# its own PolicyEngine check above, so the tool-level check is redundant.
ctx.tool_call_count += 1
_DIRECT_DISPATCH_TOOLS = {"read_file", "write_file", "edit_file", "shell_execute", "notebook_add"}
use_direct_dispatch = (
self._default_workspace is not None
and self.workspace != self._default_workspace
and tool_name in _DIRECT_DISPATCH_TOOLS
) or (tool_name in _DIRECT_DISPATCH_TOOLS and tool_name not in self.tools
) or (_skip_policy_shell and tool_name == "shell_execute")
if use_direct_dispatch:
try:
result = self._direct_filesystem_op(tool_name, params)
except Exception as e:
logger.error(" [%s] Direct filesystem exception: %s", step_id, e, exc_info=True)
self._record_failure(ctx, step_id, f"Tool exception: {type(e).__name__}: {e}")
self._handle_failure(step["on_failure"], step_id, ctx)
return
else:
# Standard registry dispatch
handler = self.tools.get(tool_name)
if not handler:
self._record_failure(ctx, step_id, f"Unknown tool: {tool_name}")
self._handle_failure(step["on_failure"], step_id, ctx)
return
try:
result = handler(params)
except Exception as e:
logger.error(" [%s] Tool exception: %s", step_id, e, exc_info=True)
self._record_failure(ctx, step_id, f"Tool exception: {type(e).__name__}: {e}")
self._handle_failure(step["on_failure"], step_id, ctx)
return
# Coerce result to string
if isinstance(result, dict):
if result.get("status") == "error":
self._record_failure(ctx, step_id, f"Tool error: {result.get('error', 'Unknown')}")
self._handle_failure(step["on_failure"], step_id, ctx)
return
result_str = json.dumps(result, default=str)
else:
result_str = str(result)
# Bind result
if "bind_result" in step:
ctx.bindings[step["bind_result"]] = result_str
# Run validation checks
passed = self._run_validation(step["validation"], result_str, ctx)
ctx.step_results[step_id] = {
"status": "pass" if passed else "fail",
"tool": tool_name,
"result_preview": result_str[:500],
"validation_passed": passed,
}
if not passed:
logger.warning(" [%s] Validation failed", step_id)
self._handle_failure(step["on_failure"], step_id, ctx)
# Ingest into worker substrate + record outcome
try:
from worker_ng import ingest_tool_result, record_tool_outcome
ingest_tool_result(self.ng, tool_name, params, result_str)
record_tool_outcome(
self.ng, tool_name, step_id, passed,
strength=1.0, context=result_str[:200],
)
except Exception as e:
logger.warning(" [%s] Substrate ingestion failed: %s", step_id, e)
# ------------------------------------------------------------------
# Gate steps
# ------------------------------------------------------------------
def _execute_gate(self, step: dict, ctx: ExecutionContext, _constraints: dict):
gate_type = step["gate_type"]
step_id = step["id"]
logger.info(" [%s] Gate: %s (%s)", step_id, step["description"], gate_type)
if gate_type == "auto_approve":
ctx.step_results[step_id] = {"status": "auto_approved", "gate_type": gate_type}
return
# For human_review and qb_checkpoint, record as pending.
# The execution report captures the checkpoint state.
# QB or human resumes by re-submitting from the next step.
ctx.step_results[step_id] = {
"status": f"pending_{gate_type}",
"gate_type": gate_type,
"staged_actions": step.get("staged_actions", []),
"description": step["description"],
}
# Gates don't abort β they pause. QB decides whether to continue.
# For now, we continue execution past gates (QB reviews the report after).
# ------------------------------------------------------------------
# Condition steps
# ------------------------------------------------------------------
def _execute_condition(self, step: dict, ctx: ExecutionContext, constraints: dict):
step_id = step["id"]
check_result = self._evaluate_check(step["check"], None, ctx)
branch = step["if_true"] if check_result else step["if_false"]
logger.info(" [%s] Condition β %s", step_id, "if_true" if check_result else "if_false")
ctx.step_results[step_id] = {
"status": "evaluated",
"branch_taken": "if_true" if check_result else "if_false",
}
for sub_step in branch:
if ctx.aborted:
break
self._execute_step(sub_step, ctx, constraints)
# ------------------------------------------------------------------
# Loop steps
# ------------------------------------------------------------------
def _execute_loop(self, step: dict, ctx: ExecutionContext, constraints: dict):
step_id = step["id"]
over = step["over"]
bind_item = step.get("bind_item", "$item")
max_iter = step.get("max_iterations", 20)
# Resolve items
if "items" in over:
items = over["items"]
elif "from_result" in over:
raw = ctx.bindings.get(over["from_result"], "")
delimiter = over.get("split_on", "\n")
items = [i.strip() for i in raw.split(delimiter) if i.strip()]
else:
items = []
items = items[:max_iter]
logger.info(" [%s] Loop over %d items", step_id, len(items))
for item in items:
if ctx.aborted:
break
ctx.bindings[bind_item] = item
for sub_step in step["body"]:
if ctx.aborted:
break
self._execute_step(sub_step, ctx, constraints)
ctx.step_results[step_id] = {"status": "completed", "iterations": len(items)}
# ------------------------------------------------------------------
# Group steps
# ------------------------------------------------------------------
def _execute_group(self, step: dict, ctx: ExecutionContext, constraints: dict):
step_id = step["id"]
logger.info(" [%s] Group: %s", step_id, step.get("description", ""))
for sub_step in step["steps"]:
if ctx.aborted:
break
self._execute_step(sub_step, ctx, constraints)
group_ok = all(
ctx.step_results.get(s["id"], {}).get("status") != "fail"
for s in step["steps"]
if s["id"] in ctx.step_results
)
ctx.step_results[step_id] = {"status": "pass" if group_ok else "fail"}
if not group_ok and "on_failure" in step:
self._handle_failure(step["on_failure"], step_id, ctx)
# ------------------------------------------------------------------
# Direct filesystem operations (workspace override bypass)
# ------------------------------------------------------------------
def _direct_filesystem_op(self, tool_name: str, params: dict):
"""Execute filesystem tools directly against self.workspace.
Bypasses TOOL_REGISTRY (and therefore FilesystemTool._check_path)
when the spec's workspace differs from the default repo path.
The executor already ran PolicyEngine checks with the correct workspace.
"""
if tool_name == "read_file":
path = params.get("path", "")
target = (self.workspace / path).resolve()
if not target.is_relative_to(self.workspace.resolve()):
return {"status": "error", "error": f"Path escapes workspace: {path}"}
start = params.get("start_line")
end = params.get("end_line")
content = target.read_text(encoding="utf-8", errors="ignore")
lines = content.splitlines()
if start is not None and end is not None:
lines = lines[max(0, start - 1):end]
return "\n".join(lines)
elif tool_name == "write_file":
path = params.get("path", "")
content = params.get("content", "")
target = (self.workspace / path).resolve()
if not target.is_relative_to(self.workspace.resolve()):
return {"status": "error", "error": f"Path escapes workspace: {path}"}
target.parent.mkdir(parents=True, exist_ok=True)
target.write_text(content, encoding="utf-8")
size = target.stat().st_size
return f"Written to {target} ({size:,} bytes)"
elif tool_name == "edit_file":
path = params.get("path", "")
old_text = params.get("old_text", "")
new_text = params.get("new_text", "")
target = (self.workspace / path).resolve()
if not target.is_relative_to(self.workspace.resolve()):
return {"status": "error", "error": f"Path escapes workspace: {path}"}
content = target.read_text(encoding="utf-8", errors="ignore")
count = content.count(old_text)
if count == 0:
return {"status": "error", "error": f"old_text not found in {path}"}
if count > 1:
return {"status": "error", "error": f"old_text found {count} times in {path} β must be unique"}
new_content = content.replace(old_text, new_text, 1)
target.write_text(new_content, encoding="utf-8")
return new_content
elif tool_name == "shell_execute":
import subprocess as _sp
command = params.get("command", "")
if not command:
return {"status": "error", "error": "Empty command"}
result = _sp.run(
command, shell=True, capture_output=True, text=True,
cwd=str(self.workspace), timeout=120,
)
return (result.stdout + result.stderr).strip()
elif tool_name == "notebook_add":
content = params.get("content", "")
logger.info(" [notebook_add] %s", content[:120])
return f"noted: {content[:60]}"
return {"status": "error", "error": f"No direct dispatch for {tool_name}"}
# ------------------------------------------------------------------
# Validation
# ------------------------------------------------------------------
def _run_validation(self, validation: dict, result: str, ctx: ExecutionContext) -> bool:
for check in validation["checks"]:
if not self._evaluate_check(check, result, ctx):
desc = check.get("description", check["operator"])
logger.warning(" Validation check failed: %s", desc)
return False
return True
def _evaluate_check(self, check: dict, default_target: Optional[str], ctx: ExecutionContext) -> bool:
# Resolve target
target_ref = check.get("target")
if target_ref and target_ref.startswith("$"):
target = ctx.bindings.get(target_ref, "")
else:
target = default_target or ""
op = check["operator"]
value = check.get("value", "")
if op == "contains":
return str(value) in target
elif op == "not_contains":
return str(value) not in target
elif op == "equals":
return target.strip() == str(value).strip()
elif op == "not_equals":
return target.strip() != str(value).strip()
elif op == "matches_regex":
return bool(re.search(str(value), target))
elif op == "result_is_string":
return isinstance(target, str)
elif op == "result_is_not_error":
return "Error:" not in target and '"status": "error"' not in target
elif op == "file_exists":
return (self.workspace / str(value)).exists()
elif op == "file_contains":
# value format: "path::substring"
parts = str(value).split("::", 1)
if len(parts) == 2:
try:
content = (self.workspace / parts[0]).read_text(errors="ignore")
return parts[1] in content
except OSError:
return False
return False
elif op == "output_length_gt":
return len(target) > int(value)
elif op == "output_length_lt":
return len(target) < int(value)
logger.warning(" Unknown operator: %s β failing safe", op)
return False
# ------------------------------------------------------------------
# Bindings
# ------------------------------------------------------------------
def _resolve_bindings(self, params: dict, ctx: ExecutionContext) -> dict:
"""Replace $var references in param values with bound results."""
resolved = {}
for k, v in params.items():
if isinstance(v, str) and v.startswith("$"):
resolved[k] = ctx.bindings.get(v, v)
else:
resolved[k] = v
return resolved
# ------------------------------------------------------------------
# Failure handling
# ------------------------------------------------------------------
def _handle_failure(self, handler, step_id: str, ctx: ExecutionContext):
# Normalize string shorthand (e.g. "abort", "skip") to full dict form.
if isinstance(handler, str):
_shorthand_map = {
"abort": "abort_block",
"abort_block": "abort_block",
"continue": "skip",
"skip": "skip",
"retry": "retry",
"goto": "goto",
"escalate_to_qb": "escalate_to_qb",
}
handler = {"action": _shorthand_map.get(handler, "abort_block")}
action = handler["action"]
msg = handler.get("message", f"Step {step_id} failed")
if action == "abort_block":
ctx.aborted = True
ctx.abort_reason = msg
logger.error(" ABORT: %s", msg)
elif action == "retry":
# Retry is handled by the caller re-invoking _execute_action
# For now, log it β real retry needs a loop wrapper
logger.info(" RETRY requested for %s (not yet implemented in executor)", step_id)
elif action == "skip":
logger.info(" SKIP: %s β continuing", msg)
elif action == "goto":
# goto requires step index lookup β deferred
logger.info(" GOTO %s requested (not yet implemented)", handler.get("goto_step"))
elif action == "escalate_to_qb":
ctx.aborted = True
ctx.abort_reason = f"ESCALATED: {msg}"
logger.warning(" ESCALATE TO QB: %s", msg)
def _record_failure(self, ctx: ExecutionContext, step_id: str, reason: str):
ctx.step_results[step_id] = {"status": "fail", "reason": reason}
# ------------------------------------------------------------------
# Report
# ------------------------------------------------------------------
def _build_report(self, spec: dict, ctx: ExecutionContext) -> dict:
"""Build the execution report for QB."""
total = len(ctx.step_results)
passed = sum(1 for r in ctx.step_results.values() if r.get("status") == "pass")
failed = sum(1 for r in ctx.step_results.values() if r.get("status") == "fail")
pending = sum(1 for r in ctx.step_results.values() if "pending" in r.get("status", ""))
return {
"report_version": "1.0.0",
"block_id": spec["block"]["id"],
"block_name": spec["block"]["name"],
"agent": spec["block"].get("agent", "unnamed"),
"status": "aborted" if ctx.aborted else ("complete" if failed == 0 else "partial_failure"),
"abort_reason": ctx.abort_reason if ctx.aborted else None,
"summary": {
"total_steps": total,
"passed": passed,
"failed": failed,
"pending_review": pending,
"tool_calls": ctx.tool_call_count,
"elapsed_seconds": round(time.time() - ctx.start_time, 2),
},
"step_results": ctx.step_results,
"bindings": {k: v[:200] for k, v in ctx.bindings.items()},
"acceptance_criteria": spec["block"]["acceptance_criteria"],
}
|