Port Verify — act101 Agent Skill

Port Verify

Gate a cross-language port for correctness — composes port-scoped behavioral equivalence, contract parity, and port-manifest drift into a port-correctness verdict. Use to verify that a ported function matches its source before marking the port done.

port-verify

A port-correctness gate: confirm a ported function behaves like its source and that the port manifest is not drifting. Composes two Enterprise verification ops with the port inventory.

Honesty caveat (read first)

By default this is a structural/contract comparison. verify_port_parity also offers opt-in differential execution (execute: true) — it generates inputs from the source signature and runs both functions under a resource-capped subprocess, diffing their JSON outputs; eligible only when both sides run under a supported interpreter (node, python3) with a JSON-able signature, falling back to structural comparison otherwise. Structurally, verify_port_parity and verify_behavioral_equivalence (scope=port) compare only dimensions BOTH grammars model — a dimension modeled by one side only is reported unknown, never falsely preserved/diverged. Each result carries modeled_kinds; an empty set for either grammar means that dimension is uncovered. Per-grammar degradation across the source/target language pair is the norm — name both languages and the uncovered dimensions. Verdicts are advisory.

Tier

Enterprise. verify_behavioral_equivalence at scope:"port", verify_port_parity, and the port_* manifest/inventory tools are all Enterprise and enforce the tier themselves.

Tools (in order)

Step	Tool	What it answers
1	`verify_behavioral_equivalence` (`scope:"port"`)	Do the source and ported functions have an equivalent control-flow shape under cross-language normalization?
2	`verify_port_parity`	Do signature arity, return presence, effect-kind set, CFG shape, and raise count match across the language pair?
3	`port_inventory`	Is the manifest consistent — is this symbol marked ported, and is the target verified? Surfaces drift.

Workflow

Call verify_behavioral_equivalence with scope:"port", the source target/file and the ported version via before/after (look up the source→target mapping with port_inventory first). equivalent confirms shape held; changed lists which dimensions diverged; unknown means a dimension could not be normalized across the pair — do not claim parity on it.
Call verify_port_parity with source_file/source_target and ported_file/ported_target. Read the verdict (preserved/diverged/ unknown), the dimensions_checked, and every mismatch. A diverged dimension is a hard gate failure.
Call port_inventory to check manifest state for the file: is it marked ported, are ported/stubbed symbols recorded, does target verification pass? A symbol that is structurally equivalent but absent/stubbed in the manifest is drift — the manifest and reality disagree.

Verdict synthesis

PORT VERIFIED — equivalent (port scope) AND parity preserved with ≥1 jointly-modeled dimension AND the manifest marks the symbol ported with a verified target.
PORT DIVERGED — equivalence changed or parity diverged on any modeled dimension; report the exact dimension and the mismatch.
MANIFEST DRIFT — structurally equivalent but the manifest disagrees (unrecorded, stubbed, or unverified target).
UNKNOWN — either op returned unknown/empty modeled_kinds on a dimension; name the source+target languages and the uncovered dimension. Never present UNKNOWN as VERIFIED.

Output

Per ported symbol: the equivalence result, the parity verdict with dimensions_checked + mismatches, the manifest status, and the gate verdict. Quote modeled_kinds for both languages on any UNKNOWN.