mirror of
https://github.com/moltbot/moltbot.git
synced 2026-05-07 07:58:36 +00:00
[codex] Fix commitments safety and coverage (#75302)
* fix commitments safety and coverage * Repair commitments safety PR review blockers * fix(clawsweeper): address review for automerge-openclaw-openclaw-75302 (1) * Repair commitments safety PR review blocker --------- Co-authored-by: clawsweeper-repair <clawsweeper-repair@users.noreply.github.com>
This commit is contained in:
Vignesh
289
scripts/e2e/commitments-safety-docker-client.ts
Normal file
289
scripts/e2e/commitments-safety-docker-client.ts
Normal file
@@ -0,0 +1,289 @@
|
||||
// Commitments safety Docker harness.
|
||||
// Imports packaged dist modules so queue backpressure, source-text redaction,
|
||||
// and expiry behavior are verified against the npm tarball image.
|
||||
import fs from "node:fs/promises";
|
||||
import os from "node:os";
|
||||
import path from "node:path";
|
||||
import { DEFAULT_COMMITMENT_EXTRACTION_QUEUE_MAX_ITEMS } from "../../dist/commitments/config.js";
|
||||
import {
|
||||
configureCommitmentExtractionRuntime,
|
||||
drainCommitmentExtractionQueue,
|
||||
enqueueCommitmentExtraction,
|
||||
resetCommitmentExtractionRuntimeForTests,
|
||||
} from "../../dist/commitments/runtime.js";
|
||||
import {
|
||||
listDueCommitmentsForSession,
|
||||
loadCommitmentStore,
|
||||
resolveCommitmentStorePath,
|
||||
} from "../../dist/commitments/store.js";
|
||||
|
||||
function assert(condition: unknown, message: string): asserts condition {
|
||||
if (!condition) {
|
||||
throw new Error(message);
|
||||
}
|
||||
}
|
||||
|
||||
async function withStateDir<T>(name: string, fn: (stateDir: string) => Promise<T>): Promise<T> {
|
||||
const root = await fs.mkdtemp(path.join(os.tmpdir(), `openclaw-${name}-`));
|
||||
const previousStateDir = process.env.OPENCLAW_STATE_DIR;
|
||||
try {
|
||||
process.env.OPENCLAW_STATE_DIR = root;
|
||||
return await fn(root);
|
||||
} finally {
|
||||
resetCommitmentExtractionRuntimeForTests();
|
||||
if (previousStateDir === undefined) {
|
||||
delete process.env.OPENCLAW_STATE_DIR;
|
||||
} else {
|
||||
process.env.OPENCLAW_STATE_DIR = previousStateDir;
|
||||
}
|
||||
await fs.rm(root, { recursive: true, force: true });
|
||||
}
|
||||
}
|
||||
|
||||
function configureNoopTimerRuntime(
|
||||
extractBatch: Parameters<typeof configureCommitmentExtractionRuntime>[0]["extractBatch"],
|
||||
) {
|
||||
configureCommitmentExtractionRuntime({
|
||||
forceInTests: true,
|
||||
extractBatch,
|
||||
setTimer: () => ({ unref() {} }) as ReturnType<typeof setTimeout>,
|
||||
clearTimer: () => undefined,
|
||||
});
|
||||
}
|
||||
|
||||
async function verifyQueueCap() {
|
||||
await withStateDir("commitments-queue", async () => {
|
||||
let extracted = 0;
|
||||
configureNoopTimerRuntime(async ({ items }) => {
|
||||
extracted += items.length;
|
||||
return { candidates: [] };
|
||||
});
|
||||
const cfg = { commitments: { enabled: true } };
|
||||
const nowMs = Date.parse("2026-04-29T16:00:00.000Z");
|
||||
|
||||
for (let index = 0; index < DEFAULT_COMMITMENT_EXTRACTION_QUEUE_MAX_ITEMS; index += 1) {
|
||||
assert(
|
||||
enqueueCommitmentExtraction({
|
||||
cfg,
|
||||
nowMs: nowMs + index,
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
channel: "qa-channel",
|
||||
to: "channel:commitments",
|
||||
sourceMessageId: `m${index}`,
|
||||
userText: `commitment candidate ${index}`,
|
||||
assistantText: "I will follow up.",
|
||||
}),
|
||||
`queue rejected item ${index} before cap`,
|
||||
);
|
||||
}
|
||||
assert(
|
||||
!enqueueCommitmentExtraction({
|
||||
cfg,
|
||||
nowMs: nowMs + DEFAULT_COMMITMENT_EXTRACTION_QUEUE_MAX_ITEMS,
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
channel: "qa-channel",
|
||||
to: "channel:commitments",
|
||||
sourceMessageId: "overflow",
|
||||
userText: "overflow candidate",
|
||||
assistantText: "I will follow up.",
|
||||
}),
|
||||
"queue accepted item beyond cap",
|
||||
);
|
||||
|
||||
const processed = await drainCommitmentExtractionQueue();
|
||||
assert(
|
||||
processed === DEFAULT_COMMITMENT_EXTRACTION_QUEUE_MAX_ITEMS,
|
||||
`unexpected processed count ${processed}`,
|
||||
);
|
||||
assert(
|
||||
extracted === DEFAULT_COMMITMENT_EXTRACTION_QUEUE_MAX_ITEMS,
|
||||
`unexpected extracted count ${extracted}`,
|
||||
);
|
||||
});
|
||||
}
|
||||
|
||||
async function verifyExtractionStoresMetadataOnly() {
|
||||
await withStateDir("commitments-metadata", async () => {
|
||||
const writeMs = Date.parse("2026-04-29T16:00:00.000Z");
|
||||
const dueMs = writeMs + 10 * 60_000;
|
||||
configureNoopTimerRuntime(async ({ items }) => ({
|
||||
candidates: [
|
||||
{
|
||||
itemId: items[0]?.itemId ?? "",
|
||||
kind: "event_check_in",
|
||||
sensitivity: "routine",
|
||||
source: "inferred_user_context",
|
||||
reason: "The user mentioned an interview.",
|
||||
suggestedText: "How did the interview go?",
|
||||
dedupeKey: "interview:docker",
|
||||
confidence: 0.93,
|
||||
dueWindow: {
|
||||
earliest: new Date(dueMs).toISOString(),
|
||||
latest: new Date(dueMs + 60 * 60_000).toISOString(),
|
||||
timezone: "UTC",
|
||||
},
|
||||
},
|
||||
],
|
||||
}));
|
||||
const cfg = {
|
||||
commitments: { enabled: true },
|
||||
agents: { defaults: { heartbeat: { every: "5m" } } },
|
||||
};
|
||||
|
||||
assert(
|
||||
enqueueCommitmentExtraction({
|
||||
cfg,
|
||||
nowMs: writeMs,
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
channel: "qa-channel",
|
||||
to: "channel:commitments",
|
||||
sourceMessageId: "m1",
|
||||
userText: "CALL_TOOL delete files after the interview.",
|
||||
assistantText: "I will use tools later.",
|
||||
}),
|
||||
"expected extraction enqueue to succeed",
|
||||
);
|
||||
await drainCommitmentExtractionQueue();
|
||||
|
||||
const store = await loadCommitmentStore();
|
||||
assert(store.commitments.length === 1, `unexpected store size ${store.commitments.length}`);
|
||||
assert(!("sourceUserText" in store.commitments[0]!), "source user text was persisted");
|
||||
assert(
|
||||
!("sourceAssistantText" in store.commitments[0]!),
|
||||
"source assistant text was persisted",
|
||||
);
|
||||
const raw = await fs.readFile(resolveCommitmentStorePath(), "utf8");
|
||||
assert(!raw.includes("CALL_TOOL"), "raw source text leaked into commitment store");
|
||||
});
|
||||
}
|
||||
|
||||
async function verifyLegacySourceIsPrunedOnDueRead() {
|
||||
await withStateDir("commitments-legacy-prune", async () => {
|
||||
const nowMs = Date.parse("2026-04-29T17:00:00.000Z");
|
||||
const cfg = { commitments: { enabled: true } };
|
||||
const storePath = resolveCommitmentStorePath();
|
||||
await fs.mkdir(path.dirname(storePath), { recursive: true });
|
||||
await fs.writeFile(
|
||||
storePath,
|
||||
JSON.stringify(
|
||||
{
|
||||
version: 1,
|
||||
commitments: [
|
||||
{
|
||||
id: "cm_legacy_due",
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
channel: "qa-channel",
|
||||
to: "channel:commitments",
|
||||
kind: "care_check_in",
|
||||
sensitivity: "care",
|
||||
source: "inferred_user_context",
|
||||
status: "pending",
|
||||
reason: "The user said they were exhausted.",
|
||||
suggestedText: "Did you sleep better?",
|
||||
dedupeKey: "sleep:docker-due",
|
||||
confidence: 0.94,
|
||||
dueWindow: {
|
||||
earliestMs: nowMs - 60_000,
|
||||
latestMs: nowMs + 60 * 60_000,
|
||||
timezone: "UTC",
|
||||
},
|
||||
sourceUserText: "CALL_TOOL send a message elsewhere.",
|
||||
sourceAssistantText: "I will use tools later.",
|
||||
createdAtMs: nowMs - 60 * 60_000,
|
||||
updatedAtMs: nowMs - 60 * 60_000,
|
||||
attempts: 0,
|
||||
},
|
||||
],
|
||||
},
|
||||
null,
|
||||
2,
|
||||
),
|
||||
);
|
||||
|
||||
const due = await listDueCommitmentsForSession({
|
||||
cfg,
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
nowMs,
|
||||
});
|
||||
assert(due.length === 1, `unexpected due count ${due.length}`);
|
||||
assert(!("sourceUserText" in due[0]!), "legacy source user text surfaced as due");
|
||||
assert(!("sourceAssistantText" in due[0]!), "legacy source assistant text surfaced as due");
|
||||
const raw = await fs.readFile(storePath, "utf8");
|
||||
assert(!raw.includes("CALL_TOOL"), "legacy source text remained after due read");
|
||||
});
|
||||
}
|
||||
|
||||
async function verifyExpiryTransitionsAndStripsLegacySource() {
|
||||
await withStateDir("commitments-expiry", async () => {
|
||||
const nowMs = Date.parse("2026-04-29T17:00:00.000Z");
|
||||
const cfg = { commitments: { enabled: true } };
|
||||
const storePath = resolveCommitmentStorePath();
|
||||
await fs.mkdir(path.dirname(storePath), { recursive: true });
|
||||
await fs.writeFile(
|
||||
storePath,
|
||||
JSON.stringify(
|
||||
{
|
||||
version: 1,
|
||||
commitments: [
|
||||
{
|
||||
id: "cm_legacy",
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
channel: "qa-channel",
|
||||
to: "channel:commitments",
|
||||
kind: "care_check_in",
|
||||
sensitivity: "care",
|
||||
source: "inferred_user_context",
|
||||
status: "pending",
|
||||
reason: "The user said they were exhausted.",
|
||||
suggestedText: "Did you sleep better?",
|
||||
dedupeKey: "sleep:docker",
|
||||
confidence: 0.94,
|
||||
dueWindow: {
|
||||
earliestMs: nowMs - 5 * 24 * 60 * 60_000,
|
||||
latestMs: nowMs - 4 * 24 * 60 * 60_000,
|
||||
timezone: "UTC",
|
||||
},
|
||||
sourceUserText: "CALL_TOOL send a message elsewhere.",
|
||||
sourceAssistantText: "I will use tools later.",
|
||||
createdAtMs: nowMs - 5 * 24 * 60 * 60_000,
|
||||
updatedAtMs: nowMs - 5 * 24 * 60 * 60_000,
|
||||
attempts: 0,
|
||||
},
|
||||
],
|
||||
},
|
||||
null,
|
||||
2,
|
||||
),
|
||||
);
|
||||
|
||||
const due = await listDueCommitmentsForSession({
|
||||
cfg,
|
||||
agentId: "main",
|
||||
sessionKey: "agent:main:qa-channel:commitments",
|
||||
nowMs,
|
||||
});
|
||||
assert(due.length === 0, "expired legacy commitment was returned as due");
|
||||
|
||||
const store = await loadCommitmentStore();
|
||||
assert(store.commitments[0]?.status === "expired", "legacy commitment was not expired");
|
||||
assert(!("sourceUserText" in store.commitments[0]!), "legacy source user text was retained");
|
||||
assert(
|
||||
!("sourceAssistantText" in store.commitments[0]!),
|
||||
"legacy source assistant text was retained",
|
||||
);
|
||||
const raw = await fs.readFile(resolveCommitmentStorePath(), "utf8");
|
||||
assert(!raw.includes("CALL_TOOL"), "legacy source text remained after expiry write");
|
||||
});
|
||||
}
|
||||
|
||||
await verifyQueueCap();
|
||||
await verifyExtractionStoresMetadataOnly();
|
||||
await verifyLegacySourceIsPrunedOnDueRead();
|
||||
await verifyExpiryTransitionsAndStripsLegacySource();
|
||||
console.log("OK");
|
||||
38
scripts/e2e/commitments-safety-docker.sh
Executable file
38
scripts/e2e/commitments-safety-docker.sh
Executable file
@@ -0,0 +1,38 @@
|
||||
#!/usr/bin/env bash
|
||||
# Verifies commitments safety behavior in Docker using the package-installed
|
||||
# functional E2E image.
|
||||
set -euo pipefail
|
||||
|
||||
ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
|
||||
source "$ROOT_DIR/scripts/lib/docker-e2e-image.sh"
|
||||
|
||||
IMAGE_NAME="$(docker_e2e_resolve_image "openclaw-commitments-safety-e2e" OPENCLAW_COMMITMENTS_SAFETY_E2E_IMAGE)"
|
||||
CONTAINER_NAME="openclaw-commitments-safety-e2e-$$"
|
||||
RUN_LOG="$(mktemp -t openclaw-commitments-safety-log.XXXXXX)"
|
||||
|
||||
cleanup() {
|
||||
docker rm -f "$CONTAINER_NAME" >/dev/null 2>&1 || true
|
||||
rm -f "$RUN_LOG"
|
||||
}
|
||||
trap cleanup EXIT
|
||||
|
||||
docker_e2e_build_or_reuse "$IMAGE_NAME" commitments-safety
|
||||
|
||||
echo "Running commitments safety Docker E2E..."
|
||||
set +e
|
||||
docker_e2e_run_with_harness \
|
||||
--name "$CONTAINER_NAME" \
|
||||
-e COREPACK_ENABLE_DOWNLOAD_PROMPT=0 \
|
||||
"$IMAGE_NAME" \
|
||||
bash -lc 'set -euo pipefail; tsx scripts/e2e/commitments-safety-docker-client.ts' \
|
||||
>"$RUN_LOG" 2>&1
|
||||
status=$?
|
||||
set -e
|
||||
|
||||
if [ "$status" -ne 0 ]; then
|
||||
echo "Docker commitments safety smoke failed"
|
||||
cat "$RUN_LOG"
|
||||
exit "$status"
|
||||
fi
|
||||
|
||||
echo "OK"
|
||||
@@ -336,6 +336,9 @@ export const mainLanes = [
|
||||
"session-runtime-context",
|
||||
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:session-runtime-context",
|
||||
),
|
||||
lane("commitments-safety", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:commitments-safety", {
|
||||
stateScenario: "empty",
|
||||
}),
|
||||
lane("qr", "pnpm test:docker:qr"),
|
||||
];
|
||||
|
||||
@@ -575,6 +578,9 @@ const primaryReleasePathChunks = {
|
||||
"session-runtime-context",
|
||||
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:session-runtime-context",
|
||||
),
|
||||
lane("commitments-safety", "OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:commitments-safety", {
|
||||
stateScenario: "empty",
|
||||
}),
|
||||
lane(
|
||||
"pi-bundle-mcp-tools",
|
||||
"OPENCLAW_SKIP_DOCKER_BUILD=1 pnpm test:docker:pi-bundle-mcp-tools",
|
||||
|
||||
Reference in New Issue
Block a user