datawhalechina/hello-agents · error · Exception

When using the 'apply' tool with theorem '{theorem_name}', t

Error message

When using the 'apply' tool with theorem '{theorem_name}', theorem parameters must be added.

What it means

Raised in the no-parameter branch of apply() when the theorem name is listed in the special_theorem set but the caller did not supply parameters. Special theorems cannot be auto-instantiated by running their GPL premises (their semantics need explicit binding), so apply() refuses to run them without explicit parameters and raises immediately.

Source

Thrown at Co-creation-projects/BitSecret-GPSAgent/src/gps/symbolic_solver.py:1325

                premise_ids.update(constraints_premise_ids)

            # add operation
            operation_id = self._add_operation(('Apply', theorem_name, theorem_paras))

            # add conclusions
            fact_id, goal_ids = self._add_conclusion(theorem_gdl, replace, premise_ids, operation_id)
            if fact_id is None:
                return f"定理'{theorem}'所有前提已满足，但添加结论失败。结论可能已经存在，或者结论未通过合法性检查。"

            self._check_goals(goal_ids)

            result = f"定理'{theorem}'执行成功，以下为问题的状态更新。\n"
            return result + self._get_update(old_fact_id, old_goal_id, old_goal_status)

        else:
            if theorem_name in special_theorem:
                msg = f"When using the 'apply' tool with theorem '{theorem_name}', theorem parameters must be added."
                raise Exception(msg)

            if ('perimeter' in theorem_name or 'area' in theorem_name or
                    'similar' in theorem_name or 'congruent' in theorem_name):
                msg = ("When the theorem name contains 'perimeter', 'area', 'similar' and 'congruent', "
                       "theorem parameters must be added.")
                raise Exception(msg)

            all_goal_ids = set()
            theorem_gdl = self.parsed_gdl['Theorems'][theorem_name]
            paras, instances, premise_ids = self._run_gpl(theorem_gdl['premises_gpl'])
            for i in range(len(instances)):
                replace = dict(zip(paras, instances[i]))

                # add operation
                theorem_paras = replace_paras(theorem_gdl['paras'], replace)
                operation_id = self._add_operation(('Apply', theorem_name, theorem_paras))

                # add conclusions

View on GitHub (pinned to 606a07d341)

Solutions

Supply explicit parameters: apply('special_theorem_name(A,B,C)') with correct uppercase letters and arity.
Check membership in special_theorem before calling and choose a different theorem or add parameters.
Update agent prompts/tool docs so the model knows these theorems are never callable bare.

Example fix

# before
solver.apply('midline')  # special theorem, no params

# after
solver.apply('midline(A,B,C)')

Defensive patterns

Strategy: validation

Validate before calling

from gps.symbolic_solver import special_theorem  # module-level set
def needs_params_bare_forbidden(theorem_name):
    return theorem_name in special_theorem

# before calling: if needs_params_bare(...): require '(A,B,...)' in the string

Try / catch

try:
    solver.apply(theorem)
except Exception as e:
    if 'special' in str(e) or "parameters must be added" in str(e):
        theorem = f"{theorem.split('(')[0]}({','.join(next_points)})"  # retry with params

Prevention

Keep special_theorem membership visible to the prompting layer so bare calls are never attempted.
Default to including parameters for every apply() call — it is always accepted.
Unit-test agent-generated theorem strings against (name known, params valid, arity matches, special => has params).

When it happens

Trigger: Calling solver.apply('some_special_theorem') with no parenthesized parameter list where 'some_special_theorem' is in the special_theorem collection defined by the solver module.

Common situations: Agents trying the 'enumerate all instances' shortcut (calling apply with a bare theorem name) on a theorem that requires explicit binding; a theorem newly added to special_theorem while tool prompts still suggest bare invocation.

Related errors

AI-assisted analysis of datawhalechina/hello-agents@606a07d341 (2026-08-14). Data as JSON: /api/errors/f496a3bdd0d56222. Report an issue: GitHub.