Refuted findings 11
Candidates the review stage examined and rejected, kept for transparency.
Credential access
criticalRefutedlow confidenceThe skill's agent.py dump_keychain() runs Objection's 'ios keychain dump' against the target app via subprocess, extracting keychain credential items on the connected device, which constitutes credential access if used on an unauthorized target.
return run_objection(
Why it was refuted
dump_keychain() executes the skill's declared keychain-inspection capability ('inspect keychain') scoped to the target app during authorized assessment, so harm and undeclared fail.
Credential access
criticalRefutedlow confidenceThe skill's agent.py dump_cookies() runs Objection's 'ios cookies get' to harvest the target app's HTTP cookie store, a credential-access capability not named in the declared description.
return run_objection(
Why it was refuted
Reading the target app's HTTP cookie store is an openly documented component of the skill's declared data-storage assessment (OWASP M2 'Check for sensitive data in insecure storage', api-reference entry), within the declared app-data inspection core capability, so harm and undeclared fail.
Credential access
criticalRefutedlow confidenceThe OWASP_MOBILE_CHECKS table drives run_owasp_assessment() to automatically execute keychain dumps, plist reads, cookie harvesting, and SSL-pinning/jailbreak bypasses against the target app in one pass.
ios keychain dump
Why it was refuted
The OWASP table merely batches the skill's declared checks (keychain dump, plist read, cookie get, pinning/jailbreak bypass) with command names recorded openly in the report, so the one-pass automation is the declared core capability and harm and undeclared fail.
Credential access
criticalRefutedlow confidenceprocess.py's search_sensitive_memory() searches the target app's live memory for password, Bearer-token, JWT, api_key, and secret patterns and records matches in the assessment report.
patterns = [
Why it was refuted
Scanning the target app's memory for credential patterns is the declared 'explore app internals at runtime' capability, instructed verbatim in SKILL.md Step 6 and documented in the api-reference, so harm and undeclared fail.
Credential access
criticalRefutedmedium confidenceprocess.py's dump_keychain() runs 'ios keychain dump', parses service/account/data values into findings, and the consolidated report containing these credential previews is written to a local JSON file.
output = self._run_objection_command(
Why it was refuted
dump_keychain() performs the declared keychain-inspection capability against the target app and writes the assessment findings only to a local report file, with no egress beyond standard Objection output, so harm and undeclared fail.
Credential access
criticalRefutedlow confidenceThe skill's data-storage workflow instructs the agent to dump the target app's iOS keychain, read app environment paths, and run sqlite execute query against credential tables on the target device, harvesting credentials in a way that matches the declared authorized-assessment scope but is credential-directed and not fully declared.
sqlite execute query
Why it was refuted
The keychain dump, env path read, and SQLite credential query are the declared core capability ('inspect keychain and filesystem data ... during authorized mobile assessments'), explicitly framed by the skill itself as MASVS-STORAGE assessment of the attached target app, so the harm and undeclared gates fail on the declared-core-capability carve-out.
Code execution
criticalRefutedlow confidenceSKILL.md instructs running an SSH command as root on the jailbroken test device to launch frida-server, a remote command execution step whose authorization depends on the assessment context.
ssh root@<device_ip>
Why it was refuted
SSH to the jailbroken test device solely to start frida-server is the declared setup path for the core runtime-Frida capability, openly declared in the artifact's decision matrix ('Direct Frida server + Objection attach'), so harm and undeclared fail.
Credential access
criticalRefutedlow confidenceSKILL.md instructs the agent to dump the iOS keychain, read the app environment, and query SQLite credential tables on the target device as part of storage-security assessment.
ios keychain dump
Why it was refuted
'ios keychain dump' is literally the declared capability ('inspect keychain and filesystem data') performed as the skill's declared data-storage assessment step on the authorized target, so harm and undeclared fail.
Privilege escalation
highRefutedlow confidenceThe bundled Frida script 'jailbreak_bypass' hooks NSFileManager fileExistsAtPath: and forces a false result for Cydia/sshd/apt paths so the target app's jailbreak checks are evaded.
if (paths.some(p => this.path.includes(p))) retval.replace(0);
Why it was refuted
The jailbreak_bypass Frida script implements exactly the declared 'validate/bypass client-side protections' capability against jailbreak detection, openly shipped and taught in Step 6/Workflow 4, so harm and undeclared fail.
Privilege escalation
highRefutedlow confidenceThe skill instructs and codes running Objection's 'ios sslpinning disable' to strip the target app's TLS certificate pinning so its traffic can be intercepted in a proxy.
return run_objection(
Why it was refuted
'ios sslpinning disable' is the declared 'validate/bypass client-side protections' capability, taught in Step 4 and documented in the api-reference as authorized traffic-interception testing, so harm and undeclared fail.
Privilege escalation
highRefutedlow confidenceThe skill instructs and codes running Objection's 'ios jailbreak disable' plus return-value overrides to defeat the target app's jailbreak detection.
return run_objection(
Why it was refuted
'ios jailbreak disable' is the declared 'bypass client-side protections' capability, openly documented in the api-reference and Workflow 4, so harm and undeclared fail.
