Files
olp/lib/ir/openai-to-ir.mjs
T
40f9453d88 fix(ir): accept OpenAI role=developer at entry surface, normalize to system (#65)
* fix(ir): accept OpenAI role=developer at entry surface, normalize to system

Hermes Agent (v0.13+) and other modern openai-completions clients (Cline,
Continue.dev) default to role=developer for the high-priority instruction
slot when the model id matches OpenAI o1/o3+ reasoning family. OLP IR's
validator rejected this with 400 IR validation failed: role must be one
of system|user|assistant|tool, got "developer".

Reproduced 2026-05-27 on PI230 Hermes v0.14.0 trying olp-codex/gpt-5.5
through PI231 OLP. olp-claude (Anthropic models) was unaffected because
Hermes sends system role for Claude models (no developer-role concept
on Anthropic side).

Fix: extend openai-to-ir.mjs normalizeRole to map developer to system at
the entry boundary. The IR canonical-four-roles invariant is preserved;
every provider plugin's role-handling stays unchanged.

Same pattern as the existing function-to-tool normalization that already
lives in normalizeRole (function role was OpenAI-deprecated). Normalize-at-
entry centralizes role-spec-evolution handling in one file rather than
bloating the IR schema and forcing every provider plugin to handle each
new role.

Why not add developer to VALID_ROLES: it would require branches in three
provider plugins (anthropic, codex, mistral) all mapping to system-style
annotation anyway, plus wider IR surface for any future OpenAI role
addition.

Tests: two new pin tests in Suite IR translation:
- developer role to system translation
- mixed-role array including developer validates cleanly

768 to 770 tests, 0 fail. Verified Hermes via olp-codex no longer hits the
IR rejection error after this fix (smoke run from PI230).

ADR 0003 Amendment 3 documents the rationale + future-role policy
(normalize-at-entry unless a role genuinely conveys
provider-distinguishable semantics).

Authority: OpenAI Responses API spec developer-role + Hermes Agent
v0.14.0 reproduction.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

* test+docs: PR #65 fold-in reviewer N2 + N3

N2 — Negative control test added: unknown role (e.g. "admin") still raises
BadRequestError. Pins that the developer-to-system normalization is the
only entry-surface escape hatch; any future widening of normalizeRole that
returns role as-is for unknown inputs will be caught by this test.

N3 — ADR 0003 Amendment 3 gains a "Cache-key impact" bullet documenting
that a developer-form and system-form request with otherwise-identical
content now share the same IR and thus the same cache key (per ADR 0005).
This is by design and matches OpenAI's own backward-compat semantics; the
note exists so a future debug session investigating "why does my developer
request hit a cache entry from an old system request" finds the answer.

Test count: 770 to 771, all pass.

Skipped reviewer N1 (URL precision) — Amendment already cites multiple
authorities including the Hermes/Cline tracking convention and live
reproduction transcript; single-URL precision is over-spec.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

---------

Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-27 18:55:11 +10:00

233 lines
8.6 KiB
JavaScript

/**
* lib/ir/openai-to-ir.mjs — OpenAI Chat Completions → IR v1.0 translation
*
* Authority: ADR 0003 § "Translation direction model" and § "Required/Optional fields"
* Entry-surface authority: OpenAI Chat Completions API
* https://platform.openai.com/docs/api-reference/chat/create
*
* This is the entry adapter — the single point where an OpenAI-shaped request
* is normalized into the IR. Provider plugins never see the OpenAI shape;
* they receive only the IR (ADR 0003 § "IR is not exposed externally").
*/
import { createHash } from 'node:crypto';
import { IR_VERSION, validateIRRequest } from './types.mjs';
// ── Custom error ──────────────────────────────────────────────────────────
export class BadRequestError extends Error {
/** @param {string} message */
constructor(message) {
super(message);
this.name = 'BadRequestError';
this.statusCode = 400;
}
}
// ── Role normalization ────────────────────────────────────────────────────
/**
* Normalize entry-surface role names → IR canonical set (system/user/assistant/tool).
*
* Per ADR 0003, IR supports exactly four roles. OpenAI's chat-completions
* spec has evolved beyond that, and we keep the IR minimal by normalizing
* at the entry boundary instead of bloating IR + every provider plugin.
*
* Current normalizations:
* - `function` → `tool` — deprecated in OpenAI chat API, replaced by tool.
* - `developer` → `system` — OpenAI o1/o3+ reasoning models accept a new
* "developer" role with similar semantics to "system" (high-priority
* instructions from the developer to the model). Providers like Hermes
* Agent and Cline default to `developer` for openai-completions calls.
* OLP-side anthropic + codex providers don't differentiate developer
* from system, so the IR canonicalizes to `system` and downstream
* translations remain unchanged.
*
* @param {string} role
* @returns {string}
*/
function normalizeRole(role) {
if (role === 'function') return 'tool';
if (role === 'developer') return 'system';
return role;
}
// ── Message translation ───────────────────────────────────────────────────
/**
* @param {object} msg - an OpenAI message object
* @returns {import('./types.mjs').IRMessage}
*/
function translateMessage(msg) {
if (!msg || typeof msg !== 'object') {
throw new BadRequestError('Each message must be an object');
}
const irMsg = {
role: normalizeRole(msg.role),
content: msg.content ?? '',
};
// content can be null for tool/function calls in OpenAI shape — normalise to ''
if (irMsg.content === null) {
irMsg.content = '';
}
if (msg.name !== undefined) {
irMsg.name = String(msg.name);
}
if (msg.tool_call_id !== undefined) {
irMsg.tool_call_id = String(msg.tool_call_id);
}
// OpenAI also uses function_call (deprecated) — treat as equivalent to tool_calls
if (Array.isArray(msg.tool_calls) && msg.tool_calls.length > 0) {
irMsg.tool_calls = msg.tool_calls.map(tc => ({
id: tc.id,
type: 'function',
function: {
name: tc.function?.name ?? '',
arguments: tc.function?.arguments ?? '',
},
}));
} else if (msg.function_call && typeof msg.function_call === 'object') {
// Deprecated OpenAI function_call field — map to single tool_calls entry.
// ID is deterministic: sha256(name + NUL + arguments).slice(0,16) so that
// two identical function_call requests produce the same IR → same cache key.
// Using Date.now() here would violate ADR 0005 invariant "same inputs → same key,
// no random, no timestamp" (F3 cold-audit finding, round-5).
const fcName = msg.function_call.name ?? '';
// Canonicalize empty-args to '{}' BEFORE hashing so the hash input matches the
// emitted IR field exactly. Otherwise `arguments: ''` and `arguments: '{}'` would
// emit identical IR but compute different ids → different cache keys for
// semantically-identical requests. Folded in per D33 F3 reviewer non-blocking #1.
const fcArgs = msg.function_call.arguments || '{}';
const fcKey = createHash('sha256')
.update(`${fcName}\0${fcArgs}`)
.digest('hex')
.slice(0, 16);
irMsg.tool_calls = [{
id: `fc-${fcKey}`, // deterministic — same input → same id
type: 'function',
function: {
name: fcName,
arguments: fcArgs,
},
}];
}
return irMsg;
}
// ── Tool definitions ──────────────────────────────────────────────────────
/**
* @param {object[]} tools - OpenAI tools array
* @returns {import('./types.mjs').IRToolDefinition[]}
*/
function translateTools(tools) {
if (!Array.isArray(tools)) return [];
return tools.map((t, i) => {
if (t.type !== 'function' || !t.function) {
throw new BadRequestError(`tools[${i}]: only type='function' tools are supported in IR v1.0`);
}
return {
type: 'function',
function: {
name: t.function.name,
...(t.function.description !== undefined && { description: t.function.description }),
...(t.function.parameters !== undefined && { parameters: t.function.parameters }),
},
};
});
}
// ── Main translator ───────────────────────────────────────────────────────
/**
* Translates an OpenAI Chat Completions request body into IR v1.0.
*
* Throws BadRequestError on validation failure.
*
* @param {object} openAIRequest - parsed JSON body from POST /v1/chat/completions
* @returns {import('./types.mjs').IRRequest}
*/
export function openAIToIR(openAIRequest) {
if (!openAIRequest || typeof openAIRequest !== 'object') {
throw new BadRequestError('Request body must be a JSON object');
}
if (!openAIRequest.model || typeof openAIRequest.model !== 'string') {
throw new BadRequestError('Request body must include a non-empty "model" string');
}
if (!Array.isArray(openAIRequest.messages) || openAIRequest.messages.length === 0) {
throw new BadRequestError('Request body must include a non-empty "messages" array');
}
/** @type {import('./types.mjs').IRRequest} */
const ir = {
irVersion: IR_VERSION,
model: openAIRequest.model,
stream: openAIRequest.stream === true,
messages: openAIRequest.messages.map((m, i) => {
try {
return translateMessage(m);
} catch (e) {
throw new BadRequestError(`messages[${i}]: ${e.message}`);
}
}),
};
// Optional numeric fields — type-coerce if needed
if (openAIRequest.max_tokens !== undefined && openAIRequest.max_tokens !== null) {
const v = Number(openAIRequest.max_tokens);
if (!Number.isInteger(v) || v <= 0) {
throw new BadRequestError('max_tokens must be a positive integer');
}
ir.max_tokens = v;
}
if (openAIRequest.temperature !== undefined && openAIRequest.temperature !== null) {
const v = Number(openAIRequest.temperature);
if (isNaN(v) || v < 0 || v > 2) {
throw new BadRequestError('temperature must be a number in [0, 2]');
}
ir.temperature = v;
}
if (openAIRequest.top_p !== undefined && openAIRequest.top_p !== null) {
const v = Number(openAIRequest.top_p);
if (isNaN(v) || v < 0 || v > 1) {
throw new BadRequestError('top_p must be a number in [0, 1]');
}
ir.top_p = v;
}
if (openAIRequest.stop !== undefined && openAIRequest.stop !== null) {
if (typeof openAIRequest.stop !== 'string' && !Array.isArray(openAIRequest.stop)) {
throw new BadRequestError('stop must be a string or array of strings');
}
ir.stop = openAIRequest.stop;
}
if (openAIRequest.tools !== undefined && openAIRequest.tools !== null) {
ir.tools = translateTools(openAIRequest.tools);
}
if (openAIRequest.tool_choice !== undefined && openAIRequest.tool_choice !== null) {
ir.tool_choice = openAIRequest.tool_choice;
}
if (openAIRequest.response_format !== undefined && openAIRequest.response_format !== null) {
ir.response_format = openAIRequest.response_format;
}
// Validate the constructed IR — belt-and-suspenders
const { valid, errors } = validateIRRequest(ir);
if (!valid) {
throw new BadRequestError(`IR validation failed: ${errors.join('; ')}`);
}
return ir;
}