Small expression languages still need explicit rules. LogicStitch is a Windows AI coding workspace for reviewing boolean-token evaluation, keeping precedence and rejected input visible instead of executing arbitrary text.
def evaluate_flat(tokens):
if not tokens or len(tokens) % 2 == 0:
raise ValueError("logic.shape")
if any(type(x) is not bool for x in tokens[::2]):
raise ValueError("logic.value")
if any(x not in ("AND", "OR") for x in tokens[1::2]):
raise ValueError("logic.operator")
terms, term = [], tokens[0]
for i in range(1, len(tokens), 2):
if tokens[i] == "AND":
term = term and tokens[i + 1]
else:
terms.append(term)
term = tokens[i + 1]
return any([*terms, term])Check True OR False AND False and reject empty input, misplaced operators, parentheses or non-boolean operands without executing strings.
Compare actual behavior with the expected result. If a check fails, investigate it before applying more changes.
Base URL, credentials and a reachable server.
def evaluate_flat(tokens):
if not tokens or len(tokens) % 2 == 0:
raise ValueError("logic.shape")
if any(type(x) is not bool for x in tokens[::2]):
raise ValueError("logic.value")
if any(x not in ("AND", "OR") for x in tokens[1::2]):
raise ValueError("logic.operator")
terms, term = [], tokens[0]
for i in range(1, len(tokens), 2):
if tokens[i] == "AND":
term = term and tokens[i + 1]
else:
terms.append(term)
term = tokens[i + 1]
return any([*terms, term])LogicStitch is a Windows AI coding workspace for developers who need to reason about conditional behavior in an existing project. Open the folder and name the decision that produces an unexpected result. A connected model can help inspect the relevant code and discuss a focused revision, while the conversation keeps the expected behavior visible during your review.
The boolean_tokens.py example suggests a task involving flat Boolean expressions. In your application, ask whether the accepted input, operator precedence and failure behavior match what the caller expects. Provide a few expressions with known outcomes before requesting edits. That makes the example useful as project context without turning the homepage into a language tutorial or implying that a rule engine is supplied with the software.
An unfiltered AI model is selectable when available through your supported OpenAI-compatible or Anthropic endpoint. Configure the address and credentials it requires, and consider the model's actual context and tool support. After examining the patch, test expressions that distinguish the intended behavior from the old one, along with malformed inputs. Keep changes to expression semantics separate from formatting cleanup. The resulting review should explain which decisions changed, why they changed and how callers will observe them.
Use an expression mixing the supported operators whose result differs under the competing interpretations. Write down the intended outcome and why the application expects it. Ask the model to trace the current code before proposing a semantics change.
Only if that is part of the requested contract. Adding syntax changes the scope and may require a different parser and new error handling. Keep an existing flat-expression repair bounded unless the project explicitly needs a broader language.
Include missing operands, unknown operators and values of the wrong type. Verify how the caller receives each failure. A helper raising an error is useful only if the surrounding application handles that error according to its own contract.
It can describe apparent behavior but cannot establish unstated intent. Provide the rule's purpose and accepted outcomes, then compare those requirements with the implementation. Mark ambiguous cases for a decision rather than accepting a plausible interpretation as policy.
LogicStitch · Updated 2026-10-25