appback-remoteagent 0.23.4 → 0.23.5
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1 -1
- package/dist/bot.js +10 -165
- package/docs/RESPONSE_ROUTING.md +20 -0
- package/package.json +1 -1
- package/scripts/selftest-telegram-update.mjs +60 -25
package/README.md
CHANGED
|
@@ -111,7 +111,7 @@ Current command surface implemented in `src/bot.ts`:
|
|
|
111
111
|
| `/sandbox [codex <mode>]` | Lists Codex sandbox choices or changes the current session sandbox |
|
|
112
112
|
| `/option retry <count>` | Sets the automatic continuation turn limit and persists it to `~/.remoteagent/.env` |
|
|
113
113
|
| `/option timeout <seconds>` | Sets the provider execution timeout and persists it to `~/.remoteagent/.env` |
|
|
114
|
-
| `/option intent <count>` |
|
|
114
|
+
| `/option intent <count>` | Retired compatibility command; explains that untagged replies are delivered without content-based retries. |
|
|
115
115
|
| `/state` | Shows the session ledger that is injected as provider context |
|
|
116
116
|
| `/state clear` | Clears the current session ledger without deleting the session |
|
|
117
117
|
| `/state note <text>` | Adds an operator note to the session ledger |
|
package/dist/bot.js
CHANGED
|
@@ -32,7 +32,7 @@ const HELP_TEXT = [
|
|
|
32
32
|
"/stop",
|
|
33
33
|
"/sandbox codex <read-only|workspace-write|danger-full-access>",
|
|
34
34
|
"/status",
|
|
35
|
-
"/option [retry <count>|timeout <seconds>|
|
|
35
|
+
"/option [retry <count>|timeout <seconds>|reasoning <low|medium|high|xhigh|max>|command-menu <on|off|refresh>]",
|
|
36
36
|
"/state [clear|note <text>]",
|
|
37
37
|
"/artifacts list|cleanup <days>",
|
|
38
38
|
"/cleanup",
|
|
@@ -649,7 +649,6 @@ ${bridge.formatStatus(mapping)}`);
|
|
|
649
649
|
actionButton(ctx, "Timeout", { kind: "option.show", option: "timeout" }),
|
|
650
650
|
],
|
|
651
651
|
[
|
|
652
|
-
actionButton(ctx, "Intent", { kind: "option.show", option: "intent" }),
|
|
653
652
|
actionButton(ctx, "Reasoning", { kind: "option.show", option: "reasoning" }),
|
|
654
653
|
actionButton(ctx, "Command menu", { kind: "option.show", option: "command-menu" }),
|
|
655
654
|
],
|
|
@@ -657,9 +656,11 @@ ${bridge.formatStatus(mapping)}`);
|
|
|
657
656
|
return;
|
|
658
657
|
}
|
|
659
658
|
if (option !== "retry" && option !== "timeout" && option !== "intent" && option !== "reasoning" && option !== "command-menu") {
|
|
660
|
-
await reply(ctx,
|
|
661
|
-
|
|
662
|
-
|
|
659
|
+
await reply(ctx, formatRuntimeOptions());
|
|
660
|
+
return;
|
|
661
|
+
}
|
|
662
|
+
if (option === "intent") {
|
|
663
|
+
await reply(ctx, "The intent option has been retired. Untagged replies are delivered without content-based retries.");
|
|
663
664
|
return;
|
|
664
665
|
}
|
|
665
666
|
if (!value) {
|
|
@@ -708,9 +709,7 @@ ${bridge.formatStatus(mapping)}`);
|
|
|
708
709
|
if (!Number.isInteger(parsed) || parsed < 0 || String(parsed) !== value.trim()) {
|
|
709
710
|
await reply(ctx, option === "retry"
|
|
710
711
|
? "Invalid retry count. Use `0` or a positive integer, for example `/option retry 6`."
|
|
711
|
-
: option
|
|
712
|
-
? "Invalid intent retry count. Use `0` or a positive integer, for example `/option intent 4`."
|
|
713
|
-
: "Invalid timeout. Use seconds as a positive integer, for example `/option timeout 600`.", {
|
|
712
|
+
: "Invalid timeout. Use seconds as a positive integer, for example `/option timeout 600`.", {
|
|
714
713
|
parse_mode: "Markdown",
|
|
715
714
|
});
|
|
716
715
|
return;
|
|
@@ -722,13 +721,6 @@ ${bridge.formatStatus(mapping)}`);
|
|
|
722
721
|
await reply(ctx, `Set automatic continuation retry limit to ${formatRetryLimit(parsed)}.\n\nSaved: TELEGRAM_AUTO_PROGRESS_MAX_TURNS=${parsed}`);
|
|
723
722
|
return;
|
|
724
723
|
}
|
|
725
|
-
if (option === "intent") {
|
|
726
|
-
config.telegramUntaggedIntentRetries = parsed;
|
|
727
|
-
await upsertInstalledEnvValue("TELEGRAM_UNTAGGED_INTENT_RETRIES", String(parsed));
|
|
728
|
-
await bridge.logSystem(botId, chatId, `Runtime option TELEGRAM_UNTAGGED_INTENT_RETRIES set to ${parsed}.`);
|
|
729
|
-
await reply(ctx, `Set untagged intent retry limit to ${formatRetryLimit(parsed)}.\n\nSaved: TELEGRAM_UNTAGGED_INTENT_RETRIES=${parsed}`);
|
|
730
|
-
return;
|
|
731
|
-
}
|
|
732
724
|
if (parsed < 10) {
|
|
733
725
|
await reply(ctx, "Invalid timeout. Use at least 10 seconds, for example `/option timeout 600`.", {
|
|
734
726
|
parse_mode: "Markdown",
|
|
@@ -1639,11 +1631,8 @@ async function routeTelegramWorkLoop(bridge, botId, chatId, message, label, botM
|
|
|
1639
1631
|
const emptyResponseRetries = config.telegramEmptyResponseRetries;
|
|
1640
1632
|
const retryableErrorRetries = config.telegramRetryableErrorRetries;
|
|
1641
1633
|
const retryableErrorDelayMs = config.telegramRetryableErrorDelayMs;
|
|
1642
|
-
const untaggedIntentRetries = config.telegramUntaggedIntentRetries;
|
|
1643
1634
|
let emptyResponseRetryCount = 0;
|
|
1644
1635
|
let retryableErrorCount = 0;
|
|
1645
|
-
let untaggedIntentRetryCount = 0;
|
|
1646
|
-
let missingEvidenceRetryCount = 0;
|
|
1647
1636
|
let deliveredProgressCount = 0;
|
|
1648
1637
|
let providerCompleted = false;
|
|
1649
1638
|
const streamedProgressKeys = new Set();
|
|
@@ -1705,23 +1694,12 @@ async function routeTelegramWorkLoop(bridge, botId, chatId, message, label, botM
|
|
|
1705
1694
|
emptyResponseRetryCount = 0;
|
|
1706
1695
|
retryableErrorCount = 0;
|
|
1707
1696
|
if (parsed.kind === "progress") {
|
|
1708
|
-
untaggedIntentRetryCount = 0;
|
|
1709
|
-
missingEvidenceRetryCount = 0;
|
|
1710
1697
|
const key = progressDeliveryKey(parsed.chunks);
|
|
1711
1698
|
if (!streamedProgressKeys.has(key)) {
|
|
1712
1699
|
streamedProgressKeys.add(key);
|
|
1713
1700
|
deliveredProgressCount += 1;
|
|
1714
1701
|
if (currentSession) {
|
|
1715
|
-
|
|
1716
|
-
if (progress.repeated) {
|
|
1717
|
-
const repeatedMessage = [
|
|
1718
|
-
"Repeated progress detected. The same work pattern has appeared 3 or more times.",
|
|
1719
|
-
"Automatic continuation stopped so the task can be inspected instead of looping.",
|
|
1720
|
-
].join("\n");
|
|
1721
|
-
await bridge.logSystem(botId, chatId, repeatedMessage);
|
|
1722
|
-
autoContinue.clear(botId, chatId, sessionId);
|
|
1723
|
-
return [repeatedMessage];
|
|
1724
|
-
}
|
|
1702
|
+
await memoryService.recordProgress(currentSession.session, parsed.chunks.join("\n"));
|
|
1725
1703
|
}
|
|
1726
1704
|
await ensureStillBound(`${turnLabel} progress delivery`);
|
|
1727
1705
|
await helpers.reportProgress(parsed.chunks);
|
|
@@ -1736,26 +1714,6 @@ async function routeTelegramWorkLoop(bridge, botId, chatId, message, label, botM
|
|
|
1736
1714
|
continue;
|
|
1737
1715
|
}
|
|
1738
1716
|
if (parsed.kind === "result") {
|
|
1739
|
-
untaggedIntentRetryCount = 0;
|
|
1740
|
-
const resultText = parsed.chunks.join("\n");
|
|
1741
|
-
const evidenceIssue = classifyMissingResultEvidence(resultText);
|
|
1742
|
-
if (evidenceIssue && missingEvidenceRetryCount < 1) {
|
|
1743
|
-
missingEvidenceRetryCount += 1;
|
|
1744
|
-
const retryMessage = `${turnLabel} returned a result without required evidence: ${evidenceIssue}`;
|
|
1745
|
-
await bridge.logSystem(botId, chatId, retryMessage);
|
|
1746
|
-
prompt = appendManagedContext(formatMissingEvidenceRetryPrompt(resultText, evidenceIssue), managedContext);
|
|
1747
|
-
continue;
|
|
1748
|
-
}
|
|
1749
|
-
if (evidenceIssue) {
|
|
1750
|
-
const blockedMessage = [
|
|
1751
|
-
"Provider reported a completed result without concrete evidence after a retry.",
|
|
1752
|
-
`Reason: ${evidenceIssue}`,
|
|
1753
|
-
"Automatic continuation stopped so the work is not accepted on an unsupported claim.",
|
|
1754
|
-
].join("\n");
|
|
1755
|
-
await bridge.logSystem(botId, chatId, blockedMessage);
|
|
1756
|
-
autoContinue.clear(botId, chatId, sessionId);
|
|
1757
|
-
return [blockedMessage];
|
|
1758
|
-
}
|
|
1759
1717
|
await ensureStillBound(`${turnLabel} final delivery`);
|
|
1760
1718
|
if (currentSession) {
|
|
1761
1719
|
await memoryService.completeTask(currentSession.session, parsed.chunks.join("\n"));
|
|
@@ -1765,19 +1723,10 @@ async function routeTelegramWorkLoop(bridge, botId, chatId, message, label, botM
|
|
|
1765
1723
|
return parsed.chunks;
|
|
1766
1724
|
}
|
|
1767
1725
|
if (parsed.kind === "blocked") {
|
|
1768
|
-
untaggedIntentRetryCount = 0;
|
|
1769
|
-
missingEvidenceRetryCount = 0;
|
|
1770
1726
|
await ensureStillBound(`${turnLabel} final delivery`);
|
|
1771
1727
|
autoContinue.clear(botId, chatId, sessionId);
|
|
1772
1728
|
return parsed.chunks;
|
|
1773
1729
|
}
|
|
1774
|
-
if (looksLikeUntaggedIntentOnlyResponse(parsed.chunks.join("\n")) && untaggedIntentRetryCount < untaggedIntentRetries) {
|
|
1775
|
-
untaggedIntentRetryCount += 1;
|
|
1776
|
-
const retryMessage = `${turnLabel} returned an untagged intent-only response; asking provider to do concrete work before replying.`;
|
|
1777
|
-
await bridge.logSystem(botId, chatId, retryMessage);
|
|
1778
|
-
prompt = appendManagedContext(formatUntaggedIntentRetryPrompt(parsed.chunks.join("\n")), managedContext);
|
|
1779
|
-
continue;
|
|
1780
|
-
}
|
|
1781
1730
|
await bridge.logSystem(botId, chatId, `${turnLabel} returned an untagged response; treating it as final output.`);
|
|
1782
1731
|
autoContinue.clear(botId, chatId, sessionId);
|
|
1783
1732
|
return parsed.chunks;
|
|
@@ -1966,11 +1915,8 @@ function parseReportResponses(formattedBlocks, transform) {
|
|
|
1966
1915
|
const header = lines.slice(0, reportLineIndex).join("\n").trim();
|
|
1967
1916
|
const reportLine = lines[reportLineIndex].trim();
|
|
1968
1917
|
const match = /^REPORT:(progress|result|blocked)$/i.exec(reportLine);
|
|
1969
|
-
|
|
1918
|
+
const kind = match?.[1]?.toLowerCase() ?? "unknown";
|
|
1970
1919
|
const body = lines.slice(reportLineIndex + 1).join("\n").trim();
|
|
1971
|
-
if ((kind === "progress" || kind === "result") && looksLikeBlockedBody(body)) {
|
|
1972
|
-
kind = "blocked";
|
|
1973
|
-
}
|
|
1974
1920
|
return {
|
|
1975
1921
|
kind,
|
|
1976
1922
|
text: body ? [header, body].filter(Boolean).join("\n") : "",
|
|
@@ -2066,107 +2012,9 @@ function escapeTelegramHtml(value) {
|
|
|
2066
2012
|
function escapeTelegramHtmlAttribute(value) {
|
|
2067
2013
|
return escapeTelegramHtml(value).replace(/"/g, """);
|
|
2068
2014
|
}
|
|
2069
|
-
function looksLikeUntaggedIntentOnlyResponse(text) {
|
|
2070
|
-
const normalized = text.trim();
|
|
2071
|
-
if (!normalized) {
|
|
2072
|
-
return false;
|
|
2073
|
-
}
|
|
2074
|
-
const hasConcreteEvidence = [
|
|
2075
|
-
/REPORT:/i,
|
|
2076
|
-
/(완료|통과|실패|확인 결과|검증 결과|원인|근거|수정했습니다|배포했습니다|커밋|푸시)/,
|
|
2077
|
-
/\b(git status|git diff|npm run|node --check|docker|journalctl|grep|rg)\b/i,
|
|
2078
|
-
/`[^`]+`/,
|
|
2079
|
-
/:\d{1,5}\b/,
|
|
2080
|
-
].some((pattern) => pattern.test(normalized));
|
|
2081
|
-
if (hasConcreteEvidence) {
|
|
2082
|
-
return false;
|
|
2083
|
-
}
|
|
2084
|
-
return [
|
|
2085
|
-
/(하겠습니다|진행하겠습니다|확인하겠습니다|수정하겠습니다|검증하겠습니다|대조하겠습니다|보겠습니다)/,
|
|
2086
|
-
/(진행해서|확인해서|수정해서|검증해서).*(하겠습니다|진행하겠습니다)/,
|
|
2087
|
-
/\b(I will|I'll|I am going to|going to|will continue|will check|will verify)\b/i,
|
|
2088
|
-
].some((pattern) => pattern.test(normalized));
|
|
2089
|
-
}
|
|
2090
|
-
function formatUntaggedIntentRetryPrompt(lastResponse) {
|
|
2091
|
-
return [
|
|
2092
|
-
"The previous response did not follow the REPORT protocol and only stated intent without concrete evidence.",
|
|
2093
|
-
"Do not repeat the plan or say what you will do.",
|
|
2094
|
-
"Do concrete work now before replying again.",
|
|
2095
|
-
"Reply with exactly one first line: REPORT:progress, REPORT:result, or REPORT:blocked.",
|
|
2096
|
-
"If you cannot continue, use REPORT:blocked and state the exact blocker.",
|
|
2097
|
-
"",
|
|
2098
|
-
"Previous invalid response:",
|
|
2099
|
-
lastResponse.trim(),
|
|
2100
|
-
].join("\n");
|
|
2101
|
-
}
|
|
2102
|
-
function classifyMissingResultEvidence(text) {
|
|
2103
|
-
const normalized = text.trim();
|
|
2104
|
-
if (!normalized) {
|
|
2105
|
-
return undefined;
|
|
2106
|
-
}
|
|
2107
|
-
if (!looksLikeCompletedWorkClaim(normalized)) {
|
|
2108
|
-
return undefined;
|
|
2109
|
-
}
|
|
2110
|
-
if (hasConcreteResultEvidence(normalized)) {
|
|
2111
|
-
return undefined;
|
|
2112
|
-
}
|
|
2113
|
-
return "REPORT:result claims completed work but does not include concrete evidence.";
|
|
2114
|
-
}
|
|
2115
|
-
function looksLikeCompletedWorkClaim(text) {
|
|
2116
|
-
return [
|
|
2117
|
-
/(수정|반영|배포|커밋|푸시|전송|생성|삭제|추가|적용|구현|저장|업데이트|등록|제거|정리|마이그레이션|검증|테스트|빌드).{0,24}(완료|했습니다|됐습니다|성공|통과)/,
|
|
2118
|
-
/(완료했습니다|완료됐습니다|끝났습니다|처리했습니다)/,
|
|
2119
|
-
/\b(fixed|implemented|deployed|committed|pushed|sent|created|deleted|updated|added|removed|migrated|verified|passed|completed|built)\b/i,
|
|
2120
|
-
].some((pattern) => pattern.test(text));
|
|
2121
|
-
}
|
|
2122
|
-
function hasConcreteResultEvidence(text) {
|
|
2123
|
-
return [
|
|
2124
|
-
/```/,
|
|
2125
|
-
/`[^`]+`/,
|
|
2126
|
-
/\b[0-9a-f]{7,40}\b/i,
|
|
2127
|
-
/sha256:[0-9a-f]{20,}/i,
|
|
2128
|
-
/\b(HTTP\s+\d{3}|exit\s+\d+|active|passed|failed)\b/i,
|
|
2129
|
-
/\b(npm run|git status|git diff|node --check|docker|journalctl|curl|psql|grep|rg|bash)\b/i,
|
|
2130
|
-
/\/[A-Za-z0-9._/-]{3,}/,
|
|
2131
|
-
/\b[A-Za-z0-9._/-]+\.(?:js|ts|tsx|jsx|sql|md|json|yml|yaml|sh|py|css|html|txt|log)\b/,
|
|
2132
|
-
/:\d{1,5}\b/,
|
|
2133
|
-
/(근거|검증|변경 파일|커밋|푸시|배포|로그|명령|출력|파일|라인|경로|상태)\s*:/,
|
|
2134
|
-
].some((pattern) => pattern.test(text));
|
|
2135
|
-
}
|
|
2136
|
-
function formatMissingEvidenceRetryPrompt(lastResponse, issue) {
|
|
2137
|
-
return [
|
|
2138
|
-
"The previous REPORT:result was not accepted by RemoteAgent.",
|
|
2139
|
-
issue,
|
|
2140
|
-
"RemoteAgent does not inspect code or decide whether the work is correct.",
|
|
2141
|
-
"You, the provider, must either provide concrete evidence for the completed work or change the reply to REPORT:progress or REPORT:blocked.",
|
|
2142
|
-
"Do not repeat a bare completion claim.",
|
|
2143
|
-
"Reply with exactly one first line: REPORT:progress, REPORT:result, or REPORT:blocked.",
|
|
2144
|
-
"",
|
|
2145
|
-
"Accepted evidence examples: file paths, line references, commands and outputs, log paths, commit IDs, image digests, deployment status, or explicit verification output.",
|
|
2146
|
-
"",
|
|
2147
|
-
"Previous unsupported result:",
|
|
2148
|
-
lastResponse.trim(),
|
|
2149
|
-
].join("\n");
|
|
2150
|
-
}
|
|
2151
2015
|
function isEmptyResponseError(message) {
|
|
2152
2016
|
return /empty response|failed without any output|without stdout\/stderr/i.test(message);
|
|
2153
2017
|
}
|
|
2154
|
-
function looksLikeBlockedBody(text) {
|
|
2155
|
-
if (!text.trim()) {
|
|
2156
|
-
return false;
|
|
2157
|
-
}
|
|
2158
|
-
const blockedPatterns = [
|
|
2159
|
-
/\b(sudo|usermod|setfacl|chmod|chown|relogin|re-login|new login session)\b/i,
|
|
2160
|
-
/\b(waiting on|need you to|you need to|please run|please do|manual step|admin step|external fix)\b/i,
|
|
2161
|
-
/\b(permission denied|permission change|ssh access|api key|login required|authentication required)\b/i,
|
|
2162
|
-
/적용되면.*(다시|이어서|계속)/i,
|
|
2163
|
-
/해주시면.*(다시|이어서|계속)/i,
|
|
2164
|
-
/권한.*(필요|없)/i,
|
|
2165
|
-
/로그인 세션.*필요/i,
|
|
2166
|
-
/관리자.*조치/i,
|
|
2167
|
-
];
|
|
2168
|
-
return blockedPatterns.some((pattern) => pattern.test(text));
|
|
2169
|
-
}
|
|
2170
2018
|
function classifyRetryableProviderIssue(message, retryAfterMs) {
|
|
2171
2019
|
if (/selected model is at capacity/i.test(message)) {
|
|
2172
2020
|
return { kind: "capacity", retryAfterMs };
|
|
@@ -2531,19 +2379,16 @@ function formatRuntimeOptions() {
|
|
|
2531
2379
|
"Runtime options",
|
|
2532
2380
|
`- retry: ${formatRetryLimit(config.telegramAutoProgressMaxTurns)} (TELEGRAM_AUTO_PROGRESS_MAX_TURNS)`,
|
|
2533
2381
|
`- timeout: ${formatTimeoutSeconds(config.commandTimeoutMs)} (COMMAND_TIMEOUT_MS)`,
|
|
2534
|
-
`- intent: ${formatRetryLimit(config.telegramUntaggedIntentRetries)} (TELEGRAM_UNTAGGED_INTENT_RETRIES)`,
|
|
2535
2382
|
`- reasoning: ${getAstraReasoning()} (Astra, server-wide, CODEX_REASONING_EFFORT)`,
|
|
2536
2383
|
`- command-menu: ${config.telegramCommandMenuEnabled ? "on" : "off"} (TELEGRAM_COMMAND_MENU_ENABLED)`,
|
|
2537
2384
|
"",
|
|
2538
2385
|
"Usage:",
|
|
2539
2386
|
"/option retry <count>",
|
|
2540
2387
|
"/option timeout <seconds>",
|
|
2541
|
-
"/option intent <count>",
|
|
2542
2388
|
"/option reasoning <low|medium|high|xhigh|max>",
|
|
2543
2389
|
"/option command-menu <on|off|refresh>",
|
|
2544
2390
|
"",
|
|
2545
2391
|
"`retry 0` disables the automatic continuation limit.",
|
|
2546
|
-
"`intent 0` disables untagged intent-only response retries.",
|
|
2547
2392
|
"`command-menu refresh` reapplies Telegram slash-command autocomplete without changing the saved option.",
|
|
2548
2393
|
].join("\n");
|
|
2549
2394
|
}
|
|
@@ -2555,7 +2400,7 @@ function formatRuntimeOptionDetail(option) {
|
|
|
2555
2400
|
return `Current automatic continuation retry limit: ${formatRetryLimit(config.telegramAutoProgressMaxTurns)}\n\nUsage: \`/option retry <count>\``;
|
|
2556
2401
|
}
|
|
2557
2402
|
if (option === "intent") {
|
|
2558
|
-
return
|
|
2403
|
+
return "The intent option has been retired. Untagged replies are delivered without content-based retries.";
|
|
2559
2404
|
}
|
|
2560
2405
|
if (option === "command-menu") {
|
|
2561
2406
|
return `Current Telegram command menu: ${config.telegramCommandMenuEnabled ? "on" : "off"}\n\nUsage: \`/option command-menu on\`, \`/option command-menu off\`, or \`/option command-menu refresh\``;
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
# Provider response routing
|
|
2
|
+
|
|
3
|
+
RemoteAgent uses the explicit first-line REPORT marker to route provider replies.
|
|
4
|
+
|
|
5
|
+
- `REPORT:progress`: forward progress and continue within the configured automatic continuation limit.
|
|
6
|
+
- `REPORT:result`: forward the final reply and end the turn.
|
|
7
|
+
- `REPORT:blocked`: forward the reply and end the turn.
|
|
8
|
+
- Missing marker: forward the reply as final output.
|
|
9
|
+
|
|
10
|
+
The response body is not used to decide whether work happened, whether evidence is adequate, or whether the provider needs permission. Evidence remains a provider reporting instruction, not a delivery gate. Recording a final reply does not certify project completion.
|
|
11
|
+
|
|
12
|
+
Repeated progress text is deduplicated for delivery. It does not override the explicit status. The configured continuation limit, user stop requests, and session binding checks control further executions.
|
|
13
|
+
|
|
14
|
+
The retired `/option intent` command returns a compatibility notice. Existing `TELEGRAM_UNTAGGED_INTENT_RETRIES` values no longer trigger executions.
|
|
15
|
+
|
|
16
|
+
## Verification
|
|
17
|
+
|
|
18
|
+
Run `npm run selftest:telegram` to exercise S091 requirement corrections and negations, final replies without evidence, untagged replies, explicit progress and blocked states, queued instruction removal, timeout handling, and stopping an active progress turn.
|
|
19
|
+
|
|
20
|
+
The S091 incident was caused by body regexes matching both "정리했습니다" and "완료 보고가 아닙니다" as completion claims. Those classifiers and their corrective retry prompts have been removed rather than expanded with word exceptions.
|
package/package.json
CHANGED
|
@@ -109,6 +109,8 @@ if (persistedBotIdentity.username !== "appbackadmin_bot") {
|
|
|
109
109
|
|
|
110
110
|
const providerCalls = [];
|
|
111
111
|
let providerMode = "success";
|
|
112
|
+
let literalResponse = "";
|
|
113
|
+
let explicitResponses = [];
|
|
112
114
|
let untaggedIntentCalls = 0;
|
|
113
115
|
let missingEvidenceCalls = 0;
|
|
114
116
|
let streamingFinalProgressCalls = 0;
|
|
@@ -136,6 +138,9 @@ const provider = {
|
|
|
136
138
|
output: "REPORT:result\nusage fallback completed with evidence: `fallback-test.log`",
|
|
137
139
|
};
|
|
138
140
|
}
|
|
141
|
+
if (providerMode === "literal-result") {
|
|
142
|
+
return {provider: "codex", sessionId: request.sessionId || "mock-thread", cwd: request.cwd, output: literalResponse};
|
|
143
|
+
}
|
|
139
144
|
if (providerMode === "untagged-intent") {
|
|
140
145
|
untaggedIntentCalls += 1;
|
|
141
146
|
return {
|
|
@@ -160,7 +165,10 @@ const provider = {
|
|
|
160
165
|
: "REPORT:result\n수정 완료했습니다.\n\n근거:\n- 변경 파일: `src/example.ts`\n- 검증: `npm run check` 통과",
|
|
161
166
|
};
|
|
162
167
|
}
|
|
163
|
-
if (providerMode === "
|
|
168
|
+
if (providerMode === "explicit-status") {
|
|
169
|
+
return {provider: "codex", sessionId: request.sessionId || "mock-thread", cwd: request.cwd, output: explicitResponses.shift() ?? "REPORT:result\nunexpected extra call"};
|
|
170
|
+
}
|
|
171
|
+
if (providerMode === "queue-hold" || providerMode === "stop-hold") {
|
|
164
172
|
queueHoldStartedResolve?.();
|
|
165
173
|
await queueHoldReleasePromise;
|
|
166
174
|
return {
|
|
@@ -168,7 +176,7 @@ const provider = {
|
|
|
168
176
|
sessionId: request.sessionId || "mock-thread",
|
|
169
177
|
publicSessionId: request.publicSessionId,
|
|
170
178
|
cwd: request.cwd,
|
|
171
|
-
output: "REPORT:result\nactive queue test completed",
|
|
179
|
+
output: providerMode === "stop-hold" ? "REPORT:progress\nstop regression progress" : "REPORT:result\nactive queue test completed",
|
|
172
180
|
};
|
|
173
181
|
}
|
|
174
182
|
if (providerMode === "streaming-progress") {
|
|
@@ -359,8 +367,8 @@ if (!/^TELEGRAM_AUTO_PROGRESS_MAX_TURNS=6$/m.test(envText)) {
|
|
|
359
367
|
if (!/^COMMAND_TIMEOUT_MS=600000$/m.test(envText)) {
|
|
360
368
|
throw new Error(`Option command did not persist command timeout to .env: ${envText}`);
|
|
361
369
|
}
|
|
362
|
-
if (
|
|
363
|
-
throw new Error(
|
|
370
|
+
if (/^TELEGRAM_UNTAGGED_INTENT_RETRIES=/m.test(envText)) {
|
|
371
|
+
throw new Error("Retired intent option should not persist configuration");
|
|
364
372
|
}
|
|
365
373
|
|
|
366
374
|
const importedSecretDataDir = path.join(tmp, "imported-secret-data");
|
|
@@ -509,7 +517,7 @@ if (!calls.some((call) => call.method === "sendMessage" && /Set automatic contin
|
|
|
509
517
|
if (!calls.some((call) => call.method === "sendMessage" && /Set provider execution timeout to 600s/.test(call.text))) {
|
|
510
518
|
throw new Error(`Did not see option timeout acknowledgement. Calls: ${JSON.stringify(calls, null, 2)}`);
|
|
511
519
|
}
|
|
512
|
-
if (!calls.some((call) => call.method === "sendMessage" && /
|
|
520
|
+
if (!calls.some((call) => call.method === "sendMessage" && /intent option has been retired/.test(call.text))) {
|
|
513
521
|
throw new Error(`Did not see option intent acknowledgement. Calls: ${JSON.stringify(calls, null, 2)}`);
|
|
514
522
|
}
|
|
515
523
|
if (!calls.some((call) => call.method === "sendMessage" && /Workspace cleanup finished for S001/.test(call.text))) {
|
|
@@ -725,14 +733,8 @@ const untaggedCalls = (await fs.readFile(telegramCalls, "utf8"))
|
|
|
725
733
|
text: Buffer.from(textB64, "base64").toString("utf8"),
|
|
726
734
|
};
|
|
727
735
|
});
|
|
728
|
-
if (untaggedIntentCalls !==
|
|
729
|
-
throw new Error(
|
|
730
|
-
}
|
|
731
|
-
if (!untaggedCalls.some((call) => /untagged intent recovered/.test(call.text))) {
|
|
732
|
-
throw new Error(`Did not see recovered result after untagged intent retry. Calls: ${JSON.stringify(untaggedCalls, null, 2)}`);
|
|
733
|
-
}
|
|
734
|
-
if (untaggedCalls.some((call) => call.method === "sendMessage" && /^계속 진행해서 확인하겠습니다\.$/.test(call.text.trim()))) {
|
|
735
|
-
throw new Error(`Untagged intent-only response leaked as final Telegram message. Calls: ${JSON.stringify(untaggedCalls, null, 2)}`);
|
|
736
|
+
if (untaggedIntentCalls !== 1 || !untaggedCalls.some(call => call.text.includes("계속 진행해서 확인하겠습니다."))) {
|
|
737
|
+
throw new Error("Untagged response was not delivered in one execution");
|
|
736
738
|
}
|
|
737
739
|
|
|
738
740
|
providerMode = "missing-evidence";
|
|
@@ -752,17 +754,23 @@ const evidenceCalls = (await fs.readFile(telegramCalls, "utf8"))
|
|
|
752
754
|
text: Buffer.from(textB64, "base64").toString("utf8"),
|
|
753
755
|
};
|
|
754
756
|
});
|
|
755
|
-
if (missingEvidenceCalls !==
|
|
756
|
-
throw new Error(
|
|
757
|
-
}
|
|
758
|
-
|
|
759
|
-
|
|
760
|
-
|
|
761
|
-
|
|
762
|
-
|
|
763
|
-
|
|
764
|
-
|
|
765
|
-
|
|
757
|
+
if (missingEvidenceCalls !== 1 || !evidenceCalls.some(call => call.text.includes("수정 완료했습니다."))) {
|
|
758
|
+
throw new Error("Final response was not delivered in one execution");
|
|
759
|
+
}
|
|
760
|
+
for (const body of [
|
|
761
|
+
"플레이스미션을 누락했고, 표시 방식도 잘못 정리했습니다.\n‘더보기 목록’이 아니라 기존 넘기기 UI를 재사용하는 요구로 정정합니다.",
|
|
762
|
+
"앞선 답변은 요구사항 정정이며, 코드 수정·검증·병합 완료 보고가 아닙니다. 이번 정정으로 변경한 파일이나 커밋은 없습니다.",
|
|
763
|
+
"sudo 권한이 필요하지 않습니다. API key 변경도 없습니다.",
|
|
764
|
+
]) {
|
|
765
|
+
providerMode = "literal-result";
|
|
766
|
+
literalResponse = "REPORT:result\n" + body;
|
|
767
|
+
const countBefore = providerCalls.length;
|
|
768
|
+
await send("/batch start");
|
|
769
|
+
await send("S091 regression");
|
|
770
|
+
await send("/batch send");
|
|
771
|
+
if (providerCalls.length !== countBefore + 1) throw new Error("Result triggered extra provider execution");
|
|
772
|
+
const delivered = await readTelegramCalls();
|
|
773
|
+
if (!delivered.some(call => call.text.includes(body))) throw new Error("Result body not delivered");
|
|
766
774
|
}
|
|
767
775
|
|
|
768
776
|
providerMode = "streaming-progress";
|
|
@@ -859,6 +867,33 @@ if (providerCalls.length !== queueProviderCallsBefore + 1) {
|
|
|
859
867
|
throw new Error(`Removed queued instructions reached the provider: ${providerCalls.length - queueProviderCallsBefore} calls`);
|
|
860
868
|
}
|
|
861
869
|
|
|
870
|
+
providerMode = "explicit-status";
|
|
871
|
+
explicitResponses = ["REPORT:progress\nsudo 권한 변경은 필요 없습니다. 다음 단계를 진행합니다.", "REPORT:result\nexplicit continuation finished"];
|
|
872
|
+
const explicitBefore = providerCalls.length;
|
|
873
|
+
await send("/batch start");
|
|
874
|
+
await send("explicit status regression");
|
|
875
|
+
await send("/batch send");
|
|
876
|
+
if (providerCalls.length !== explicitBefore + 2) throw new Error("Explicit progress was overridden by body words");
|
|
877
|
+
explicitResponses = ["REPORT:blocked\n명시적으로 중단합니다."];
|
|
878
|
+
const blockedBefore = providerCalls.length;
|
|
879
|
+
await send("/batch start");
|
|
880
|
+
await send("explicit blocked regression");
|
|
881
|
+
await send("/batch send");
|
|
882
|
+
if (providerCalls.length !== blockedBefore + 1) throw new Error("Explicit blocked response retried");
|
|
883
|
+
|
|
884
|
+
providerMode = "stop-hold";
|
|
885
|
+
queueHoldStartedPromise = new Promise(resolve => { queueHoldStartedResolve = resolve; });
|
|
886
|
+
queueHoldReleasePromise = new Promise(resolve => { queueHoldReleaseResolve = resolve; });
|
|
887
|
+
const stopBefore = providerCalls.length;
|
|
888
|
+
await send("/batch start");
|
|
889
|
+
await send("stop regression");
|
|
890
|
+
const stoppedRun = send("/batch send");
|
|
891
|
+
await queueHoldStartedPromise;
|
|
892
|
+
await send("/stop");
|
|
893
|
+
queueHoldReleaseResolve();
|
|
894
|
+
await stoppedRun;
|
|
895
|
+
if (providerCalls.length !== stopBefore + 1) throw new Error("Stop allowed automatic continuation");
|
|
896
|
+
|
|
862
897
|
console.log(JSON.stringify({
|
|
863
898
|
ok: true,
|
|
864
899
|
dataDir,
|
|
@@ -867,7 +902,7 @@ console.log(JSON.stringify({
|
|
|
867
902
|
recoveredTodoItems: recoveredActive.length,
|
|
868
903
|
retryOption: 6,
|
|
869
904
|
timeoutOptionMs: 600000,
|
|
870
|
-
|
|
905
|
+
intentOptionRetired: true,
|
|
871
906
|
providerCalls: providerCalls.length,
|
|
872
907
|
untaggedIntentCalls,
|
|
873
908
|
missingEvidenceCalls,
|