Module cel 

CEL (Common Expression Language) evaluation — the ONE gated exception to the zero-dependency moat (--features cel, default OFF).

CEL is used wherever agentd evaluates a deterministic expression over run data: workflow {"op":"cel"} predicates, computed assign.expr values, infer.check value constraints, and reactive {"op":"cel"} wake conditions. Its design properties are exactly the requirements here: non-Turing-complete, no I/O, guaranteed termination — the one form of “code” a model can safely author and agentd can immediately execute.

This module is ALWAYS compiled; only its internals are feature-gated. A non-cel build answers every call with a clear “requires the ‘cel’ build feature” error, so authoring surfaces reject CEL at define/parse time (fail-closed) instead of silently mis-evaluating at run time.

Constants

FEATURE_MSG

The message every entry point returns on a build without the feature.

MAX_CEL_EXPR

Expression length cap — a routing/shaping expression is a line, not a program; an oversized one is refused at compile-check time.

Functions

compile_check

Compile-check an expression (define/parse-time validation): length cap + full CEL parse. Err carries the parser’s message for the author.

eval_bool

Evaluate an expression to a BOOLEAN with the given variables in scope (each (name, value) becomes a top-level identifier). A non-bool result is an error — a predicate must decide, not coerce.

eval_value

Evaluate an expression to a JSON VALUE with the given variables in scope — the computed-assign path.

vars_of

Convenience: a blackboard-shaped variable list (BTreeMap<String, Value> → the (name, value) slice shape the eval fns take).