Files
openclaw/src/agents/execution-contract.ts
Peter Steinberger 00d8d7ead0 refactor: extract normalization core package
Extract shared normalization/coercion helpers into private @openclaw/normalization-core workspace package while preserving existing plugin SDK helper subpaths.\n\nAlso keeps direct normalization-core imports internal, wires UI/build/loader resolution, and replaces the slow PR network CodeQL lane with a fast added-line boundary scan while retaining full CodeQL for scheduled/manual runs.\n\nVerification: local moved tests, plugin SDK boundary tests, extension loader tests, agents-support shard, UI build/test, build artifacts, lint, workflow guards, autoreview, and GitHub CI passed on PR head 963d893715.
2026-05-31 01:33:00 +01:00

124 lines
5.2 KiB
TypeScript

import { normalizeLowercaseStringOrEmpty } from "@openclaw/normalization-core/string-coerce";
import type { OpenClawConfig } from "../config/types.openclaw.js";
import { resolveAgentExecutionContract, resolveSessionAgentIds } from "./agent-scope.js";
/**
* Strip any leading `provider/` or `provider:` prefix from a model id so the
* bare-name regex matching below works against `openai/gpt-5.4` and
* `openai:gpt-5.4` the same way it does against `gpt-5.4`. Returns the bare
* model id lowercased for comparison.
*
* Without this, auto-activation silently failed on prefixed model ids — a
* user who configured `model: "openai/gpt-5.4"` in their agent config would
* get the pre-PR-H looser default behavior because the regex only matched
* bare names. The adversarial review in #64227 flagged this as a quality
* gap on completion-gate criterion 1.
*/
export function stripProviderPrefix(modelId: string): string {
const normalizedModelId = modelId.trim();
const match = /^([^/:]+)[/:](.+)$/.exec(normalizedModelId);
return (match?.[2] ?? normalizedModelId).toLowerCase();
}
/**
* Regex that matches the full set of GPT-5 variants the strict-agentic
* contract should auto-activate for. Intentionally permissive: every
* model id in the gpt-5 family should opt in by default, not just the
* canonical `gpt-5.4`.
*
* Covers:
* - `gpt-5`, `gpt-5o`, `gpt-5o-mini` (no separator after `5`)
* - `gpt-5.4`, `gpt-5.4-alt`, `gpt-5.0` (dot separator)
* - `gpt-5-preview`, `gpt-5-turbo`, `gpt-5-2025-03` (dash separator)
*
* Does NOT cover `gpt-4.5`, `gpt-6`, or any non-gpt-5 family member.
*/
const STRICT_AGENTIC_MODEL_ID_PATTERN = /^gpt-5(?:[.o-]|$)/i;
/**
* Supported provider + model combinations where strict-agentic is the intended
* runtime contract. Kept as a narrow helper so both the execution-contract
* resolver and the `update_plan` auto-enable gate converge on the same
* definition of "GPT-5-family OpenAI run". The embedded
* `mock-openai` QA lane intentionally piggybacks on that contract so repo QA
* can exercise the same incomplete-turn recovery rules end to end.
*/
export function isStrictAgenticSupportedProviderModel(params: {
provider?: string | null;
modelId?: string | null;
}): boolean {
const provider = normalizeLowercaseStringOrEmpty(params.provider ?? "");
if (provider !== "openai" && provider !== "mock-openai") {
return false;
}
const modelId = typeof params.modelId === "string" ? params.modelId : "";
const bareModelId = stripProviderPrefix(modelId);
return STRICT_AGENTIC_MODEL_ID_PATTERN.test(bareModelId);
}
/**
* Returns the effective execution contract for an embedded OpenClaw run.
*
* strict-agentic is a GPT-5-family OpenAI-only runtime contract,
* so an unsupported provider/model pair always collapses to `"default"`
* regardless of what the caller passed or what config says — the contract
* is inert off-provider. Within the supported lane, the behavior matrix is:
*
* - Supported provider/model + explicit `"strict-agentic"` in config
* (defaults or per-agent override) ⇒ `"strict-agentic"`.
* - Supported provider/model + explicit `"default"` in config ⇒ `"default"`
* (opt-out honored).
* - Supported provider/model + unspecified ⇒ `"strict-agentic"` so the
* no-stall completion-gate criterion applies to out-of-the-box GPT-5 runs
* without requiring every user to set the flag.
* - Unsupported provider/model (anything that is not openai
* with a gpt-5-family model id) ⇒ `"default"`, even when the config
* explicitly sets `"strict-agentic"`. The retry guard and blocked-exit
* helpers all check this lane again, so an explicit `"strict-agentic"`
* on an unsupported lane is a no-op rather than a hard failure.
*
* This means explicit opt-out still works, but the gate criterion
* "GPT-5.4 no longer stalls after planning" now covers unconfigured
* installations, not only users who opted in manually.
*/
export function resolveEffectiveExecutionContract(params: {
config?: OpenClawConfig;
sessionKey?: string;
agentId?: string | null;
provider?: string | null;
modelId?: string | null;
}): "default" | "strict-agentic" {
const { sessionAgentId } = resolveSessionAgentIds({
sessionKey: params.sessionKey,
config: params.config,
agentId: params.agentId ?? undefined,
});
const explicit = resolveAgentExecutionContract(params.config, sessionAgentId);
// strict-agentic is a GPT-5-family OpenAI runtime contract
// regardless of whether it was set explicitly or auto-activated. On an
// unsupported provider/model pair the contract is inert either way, so
// the effective value collapses to "default".
const supported = isStrictAgenticSupportedProviderModel({
provider: params.provider,
modelId: params.modelId,
});
if (!supported) {
return "default";
}
if (explicit === "default") {
return "default";
}
// Explicit strict-agentic OR unspecified-but-supported → strict-agentic.
return "strict-agentic";
}
export function isStrictAgenticExecutionContractActive(params: {
config?: OpenClawConfig;
sessionKey?: string;
agentId?: string | null;
provider?: string | null;
modelId?: string | null;
}): boolean {
return resolveEffectiveExecutionContract(params) === "strict-agentic";
}