Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
/**
* The transcript lane's `labeled_credential` detector (#6147 idea, widened).
* A low-entropy password typed into a conversation (`password: hunter2`,
* `login alice / hunter2`) cleared both the vendor shapes and the
* entropy-gated assignment heuristic, so it reached the searchable
* conversation page verbatim.
*
* MUST_REDACT holds every leak form from the investigation lane's bypass
* probe; MUST_KEEP holds its over-redaction cases plus code and prose a
* coding transcript is full of. Credential values are synthetic.
*/
import { describe, expect, test } from 'bun:test';
import { buildSyntheticBrainCorpus } from '../scripts/secret-scan-fp-budget.ts';
import { redactFindings, scanText } from '../src/core/secret-scan.ts';
import { LABELED_STOPLIST, labeledEchoEligible } from '../src/core/secret-scan-labeled.ts';
import { redactSession, renderSessionParts } from '../src/core/transcripts/render.ts';
import type { ParsedSession } from '../src/core/transcripts/types.ts';
const PW = ['hunter', '2'].join('');
const LONG_PW = ['Tr0ub4', 'dor&3x'].join('');
function session(texts: string[], meta: Partial<ParsedSession['meta']> = {}): ParsedSession {
return {
meta: { harness: 'claude-code', sessionId: 'labeled-cred-0001', startedAt: '2026-10-06T09:00:00.000Z', ...meta },
messages: texts.map((text, i) => ({ role: i % 2 === 0 ? 'user' : 'assistant', timestamp: `2026-10-06T09:00:0${i}.000Z`, text })),
};
}
function pageText(texts: string[]): string {
return redactSession(session(texts), { userPatternsPath: '/nonexistent' }).session.messages.map((m) => m.text).join('\n');
}
const MUST_REDACT: Array<[string, string]> = [
['bare password label', `password: ${PW}`],
['username then password on one line', `username: alice-example password: ${PW}`],
['delimiter-free login pair', `login alice-example / ${PW}`],
['dash-delimited login pair', `Login - alice-example / ${PW}`],
['single-quoted pair', `login: 'alice-example' / '${PW}'`],
['backticked pair', `login: \`alice-example\` / \`${PW}\``],
['full-width colon', `login:alice-example / ${PW}`],
['creds label', `creds: alice-example / ${PW}`],
['user/pass label', `user/pass: alice-example / ${PW}`],
['colon-joined pair', `login: alice-example:${PW}`],
['equals and slash without spaces', `login=alice-example/${PW}`],
['pair split across a newline', `login: alice-example /\n${PW}`],
['credentials pair', `credentials: alice-example / ${PW}`],
['plain login pair', `login: alice-example / ${PW}`],
['markdown bold label', `**Password:** ${PW}`],
['JSON key', `{"password": "${PW}"}`],
['prefixed env name', `DB_PASSWORD=${PW}`],
['full-width colon after password', `password:${PW}`],
['pwd label', `pwd=${PW}`],
['passcode label', `passcode: ${PW}`],
['trailing sentence punctuation', `The password: ${PW}.`],
// W4.1
['CLI flag with a separate value', `mysql -u root --password ${PW} -h db.example`],
['short CLI flag word', `psql --pass ${PW}`],
['quoted CLI value with a space', `tool --password "a b ${PW}"`],
['value alone on the next line', `password:\n${PW}`],
['digitless value on the next line', `password:\n${['hunter', 'abc'].join('')}${PW}`],
['next-line value with CRLF and indentation', `Password:\r\n ${PW}\r\n`],
['quoted next-line value (pretty JSON)', `"password":\n "${PW}",`],
['suffix after the label', `PASSWORD_DB=${PW}`],
['label inside an env name', `FOO_PASSWORD_BAR=${PW}`],
['table column', `| user | password |\n|---|---|\n| alice-example | ${PW} |`],
['table without outer pipes', `user | password\n--- | ---\nalice-example | ${PW}`],
['table cell with an escaped pipe', `| user | password |\n|---|---|\n| alice-example | a\\|${PW} |`],
['table with two credential columns', `| password | user | pwd |\n|---|---|---|\n| ${PW} | alice-example | x${PW}y |`],
['key/value table row', `| password | ${PW} |`],
['table with CRLF', `| user | password |\r\n|---|---|\r\n| alice-example | ${PW} |\r\n`],
// W12 S1: a password that starts with "-" (redacted before W4.1; only a CLI flag's value may not start with "-")
['single label, value starting with -', `password: -${PW}`],
['pair, password starting with -', `login: alice-example / -Xy9${PW}!q`],
['quoted env value starting with -', `DB_PASSWORD="-${PW}"`],
['single label, value starting with --', `password: --${PW}x9`],
// W12 S2: a pair whose user half looks like a setting suffix
['pair with a setting-like user name', `login: pass_auth / ${PW}x9`],
// W12 S4
['prefixed CLI flag', `aws rds modify-db-instance --master-user-password ${PW}x9`],
['db-password flag', `tool --db-password ${PW}x9`],
['openssl -passin', `openssl rsa -in key.pem -passin pass:${PW}x9`],
['pw flag', `client --pw ${PW}x9`],
['table with |-|-| separators', `|user|password|\n|-|-|\n|alice-example|${PW}|`],
['pw column header', `| user | pw |\n|---|---|\n| alice-example | ${PW} |`],
['qualified column header', `| user | Password (prod) |\n|---|---|\n| alice-example | ${PW} |`],
];
const MUST_KEEP: Array<[string, string]> = [
['login flow prose', 'Login: email/password flow is broken'],
['SSO provider pair', 'login: Google/GitHub SSO'],
['doc path', 'credentials: docs/auth.md'],
['host and path', 'login: example.test/path'],
['n/a', 'login: n/a'],
['type annotation', 'password: string'],
['validator chain', 'password: z.string().min(8)'],
['request field', 'password = req.body.password'],
['shell pwd output', 'pwd: /home/alice-example'],
['PWD env', 'PWD=/home/alice-example'],
['mask', 'password: ********'],
['prose after the label', 'Password: changed yesterday'],
['yes', 'password: yes'],
['login page path', 'login page / step2'],
['env reference', 'password=$DB_PASS'],
['template reference', 'password: ${DB_PASSWORD}'],
['identifier declaration', 'const hashedPassword: string'],
['identifier value', 'password: hashedPassword'],
['delimiter-free prose', 'password reset flow'],
['URL after login', 'login: https://example.test/v2'],
['placeholder', 'password: <your password>'],
['function call', 'pwd = os.getcwd()'],
['versions', 'login: v1.2/v1.3'],
['type name', 'password: SecretStr = Field()'],
['await', 'const passphrase = await passphraseFrom(args);'],
['code span label', 'the libpq form (`password=`, `sslpassword=`) is scrubbed'],
['already redacted', 'password: <REDACTED:high_entropy_assignment>'],
// W4.1
['setting suffix', 'PASSWORD_MIN_LENGTH=12345'],
['file flag', '--password-file /run/secrets/db'],
['file env', 'POSTGRES_PASSWORD_FILE=/run/secrets/db'],
['n/a table cell', '| password | n/a |'],
['policy header table', '| Password policy | Notes |\n|---|---|\n| strong | yes please |'],
['password header row is not a key/value row', '| Password | Notes |\n|---|---|\n| x | y |'],
['pass the test', 'make sure we pass the test suite'],
['prose under a password heading', 'Password:\nYou can change it in the settings page'],
['pass/fail column', '| test | pass |\n|---|---|\n| auth | ok2 |'],
// W12 S1/S4: a flag is never a flag's value; boolean and action flags are not labels
['flag followed by another flag', 'tool --password --interactive --verbose'],
['no-password boolean flag', 'ssh-keygen --no-password build-key'],
['skip-password flag', 'tool --skip-password-check release2'],
];
describe('labeled_credential on the transcript page lane', () => {
for (const [name, text] of MUST_REDACT) {
test(`redacts: ${name}`, () => {
const out = pageText([text]);
expect(out).not.toContain(PW);
expect(out).toContain('<REDACTED:labeled_credential>');
});
}
for (const [name, text] of MUST_KEEP) {
test(`keeps: ${name}`, () => {
expect(pageText([text])).toBe(text);
});
}
test('only the password half of a pair is claimed; the user name stays readable', () => {
expect(pageText([`login: alice-example / ${PW}`])).toBe('login: alice-example / <REDACTED:labeled_credential>');
});
test('the receipt counts a typed labeled_credential finding', () => {
const red = redactSession(session([`password: ${PW}`]), { userPatternsPath: '/nonexistent' });
expect(red.redactionCount).toBe(1);
expect(scanText(`password: ${PW}`, { labeledCredentials: true }).map((f) => f.pattern)).toEqual(['labeled_credential']);
});
test('a value of 8+ characters joins the session-wide echo list: a bare repeat in another message and the title is scrubbed', () => {
const red = redactSession(
session([`creds for staging: login alice-example / ${LONG_PW}`, `retrying with ${LONG_PW} now`], { title: `staging ${LONG_PW}` }),
{ userPatternsPath: '/nonexistent' },
);
expect(JSON.stringify(red.session)).not.toContain(LONG_PW);
expect(red.session.messages[1].text).toBe('retrying with <REDACTED:labeled_credential> now');
const content = renderSessionParts(red).parts[0].content;
expect(content).not.toContain(LONG_PW);
});
test('echo floor: a short or common value is redacted at its label but never scrubbed elsewhere', () => {
const out = pageText(['pwd: 1234', 'the room is 1234 and the code reviewer said 1234']);
expect(out).toBe('pwd: <REDACTED:labeled_credential>\nthe room is 1234 and the code reviewer said 1234');
expect(labeledEchoEligible('1234')).toBe(false);
expect(labeledEchoEligible('passwords')).toBe(true);
expect(labeledEchoEligible('required')).toBe(false);
expect(LABELED_STOPLIST.has('required')).toBe(true);
});
});
describe('labeled_credential is opt-in', () => {
test('off by default, and off with only highEntropy (the push gate and every other lane)', () => {
expect(scanText(`password: ${PW}`)).toEqual([]);
expect(scanText(`password: ${PW}`, { highEntropy: true })).toEqual([]);
expect(redactFindings(`login alice-example / ${PW}`, { highEntropy: true }).text).toBe(`login alice-example / ${PW}`);
});
test('the synthetic brain-like clean corpus gains no labeled_credential finding', () => {
for (const doc of buildSyntheticBrainCorpus().filter((d) => d.kind === 'clean')) {
expect(scanText(doc.text, { highEntropy: true, labeledCredentials: true }).filter((f) => f.pattern === 'labeled_credential')).toEqual([]);
}
});
});