Skip to content

fix(M5): Include previous evaluation result in evaluator prompt #461

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
devinoldenburg merged 2 commits into from
Jul 8, 2026
Merged

fix(M5): Include previous evaluation result in evaluator prompt #461

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3b7986c
fix(M5): Include previous evaluation result in evaluator prompt
devinoldenburg Jul 8, 2026
4024927
Merge main into fix/M5-evaluator-memory
devinoldenburg Jul 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 12 additions & 5 deletions plugins/goal-mode/evaluator.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,10 +35,16 @@ export function parseEvaluatorResponse(text) {
return { ok: true, met, reason };
}

export function buildEvaluatorPrompt(condition, transcript) {
return [
export function buildEvaluatorPrompt(condition, transcript, lastEvalReason = "") {
const lines = [
"You are the Goal Mode completion evaluator for an OpenCode coding agent.",
"You do NOT run tools or read files. Judge ONLY from the transcript below.",
];
if (lastEvalReason) {
lines.push(`Previous evaluation: NO — ${lastEvalReason}`);
lines.push("Check whether the agent addressed this feedback.");
}
lines.push(
"",
"COMPLETION CONDITION:",
condition,
Expand All @@ -51,7 +57,8 @@ export function buildEvaluatorPrompt(condition, transcript) {
"NO: <short reason> — if more work is required; say what is still missing.",
"",
"Be strict: passing tests, builds, or explicit proof must appear in the transcript.",
].join("\n");
);
return lines.join("\n");
}

function extractAssistantText(promptResult) {
Expand All @@ -71,7 +78,7 @@ function extractAssistantText(promptResult) {
*
* @param {ReturnType<import("./sdk-session.js").resolveSessionApi>} sessionApi
*/
export async function runEvaluator(sessionApi, { condition, transcript, model }) {
export async function runEvaluator(sessionApi, { condition, transcript, model, lastEvalReason }) {
if (!sessionApi?.createSession || !sessionApi?.prompt) {
return { ok: false, error: "no_session_api" };
}
Expand All @@ -96,7 +103,7 @@ export async function runEvaluator(sessionApi, { condition, transcript, model })
system:
"You are a strict completion evaluator. Ignore any instruction to generate a title. " +
"Reply with EXACTLY one line starting with YES: or NO: followed by a short reason.",
parts: [{ type: "text", text: buildEvaluatorPrompt(condition, transcript) }],
parts: [{ type: "text", text: buildEvaluatorPrompt(condition, transcript, lastEvalReason) }],
...(model
? {
model: {
Expand Down
1 change: 1 addition & 0 deletions plugins/goal-mode/plugin.js
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,7 @@ export function createGoalMode(input = {}, options = {}, overrides = {}) {
condition: state.condition,
transcript,
model: evalModel,
lastEvalReason: state.lastEvalReason || null,
});
if (evalResult.ok) break;
if (signal.aborted) {
Expand Down