let5see/clawdbot
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

test: harden gateway tool probes

Browse Source
This commit is contained in:
Peter Steinberger
2026-01-10 19:46:02 +00:00
parent d44bb41d27
commit afede929b3
1 changed files with 15 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
src/gateway/gateway-models.profiles.live.test.ts
View File

@@ -331,56 +331,36 @@ describeLive("gateway live (dev agent, profile keys)", () => {
if (EXTRA_TOOL_PROBES) { if (EXTRA_TOOL_PROBES) {
const nonceC = `nonceC=${randomUUID()}`; const nonceC = `nonceC=${randomUUID()}`;
const nonceD = `nonceD=${randomUUID()}`;
const toolWritePath = path.join( const toolWritePath = path.join(
tempDir, tempDir,
`write-${runIdTool}.txt`, `write-${runIdTool}.txt`,
); );
const writeProbe = await client.request<AgentFinalPayload>( const bashReadProbe = await client.request<AgentFinalPayload>(
"agent", "agent",
{ {
sessionKey, sessionKey,
idempotencyKey: `idem-${runIdTool}-write`, idempotencyKey: `idem-${runIdTool}-bash-read`,
message: message:
`Call the tool named \`write\` (or \`Write\` if \`write\` is unavailable) to write exactly "${nonceC}" to "${toolWritePath}". ` + `Call the tool named \`bash\` (or \`Bash\` if \`bash\` is unavailable) and run: ` +
`Then call the tool named \`read\` (or \`Read\`) on "${toolWritePath}". ` + `mkdir -p "${tempDir}" && printf '%s' '${nonceC}' > "${toolWritePath}" ` +
`Then call the tool named \`read\` (or \`Read\` if \`read\` is unavailable) with JSON arguments: {"path":"${toolWritePath}"} ` +
`Finally reply with exactly: ${nonceC}.`, `Finally reply with exactly: ${nonceC}.`,
deliver: false, deliver: false,
}, },
{ expectFinal: true }, { expectFinal: true },
); );
if (writeProbe?.status !== "ok") { if (bashReadProbe?.status !== "ok") {
throw new Error( throw new Error(
`write probe failed: status=${String(writeProbe?.status)}`, `bash+read probe failed: status=${String(bashReadProbe?.status)}`,
); );
} }
const writeText = extractPayloadText(writeProbe?.result); const bashReadText = extractPayloadText(bashReadProbe?.result);
if (!writeText.includes(nonceC)) { if (!bashReadText.includes(nonceC)) {
throw new Error(`write probe missing nonce: ${writeText}`);
}
const bashProbe = await client.request<AgentFinalPayload>(
"agent",
{
sessionKey,
idempotencyKey: `idem-${runIdTool}-bash`,
message:
`Call the tool named \`bash\` (or \`Bash\` if \`bash\` is unavailable) and run: echo ${nonceD}. ` +
`Then reply with exactly: ${nonceD}.`,
deliver: false,
},
{ expectFinal: true },
);
if (bashProbe?.status !== "ok") {
throw new Error( throw new Error(
`bash probe failed: status=${String(bashProbe?.status)}`, `bash+read probe missing nonce: ${bashReadText}`,
); );
} }
const bashText = extractPayloadText(bashProbe?.result);
if (!bashText.includes(nonceD)) {
throw new Error(`bash probe missing nonce: ${bashText}`);
}
await fs.rm(toolWritePath, { force: true }); await fs.rm(toolWritePath, { force: true });
} }
@@ -398,8 +378,7 @@ describeLive("gateway live (dev agent, profile keys)", () => {
{ {
sessionKey, sessionKey,
idempotencyKey: `idem-${runId2}-1`, idempotencyKey: `idem-${runId2}-1`,
message: message: `Call the tool named \`read\` (or \`Read\`) on "${toolProbePath}". Do not write any other text.`,
"Call the tool named `read` (or `Read`) on package.json. Do not write any other text.",
deliver: false, deliver: false,
}, },
{ expectFinal: true }, { expectFinal: true },
@@ -415,8 +394,7 @@ describeLive("gateway live (dev agent, profile keys)", () => {
{ {
sessionKey, sessionKey,
idempotencyKey: `idem-${runId2}-2`, idempotencyKey: `idem-${runId2}-2`,
message: message: `Now answer: what are the values of nonceA and nonceB in "${toolProbePath}"? Reply with exactly: ${nonceA} ${nonceB}.`,
'Now answer: what is the "version" field in package.json? Reply with just the version string.',
deliver: false, deliver: false,
}, },
{ expectFinal: true }, { expectFinal: true },
@@ -426,9 +404,9 @@ describeLive("gateway live (dev agent, profile keys)", () => {
`post-tool message failed: status=${String(second?.status)}`, `post-tool message failed: status=${String(second?.status)}`,
); );
} }
const version = extractPayloadText(second?.result); const reply = extractPayloadText(second?.result);
if (!/^\d{4}\.\d+\.\d+/.test(version.trim())) { if (!reply.includes(nonceA) || !reply.includes(nonceB)) {
throw new Error(`unexpected version: ${version}`); throw new Error(`unexpected reply: ${reply}`);
} }
} }
} catch (err) { } catch (err) {