agentleFS
blackfoxxxblackfoxxx/Mobix/.agents/skills/mobile-pentest/SKILL.md

mobile-pentest

Skill9 starsChanged 3 days agoMITView on GitHub

Use this skill for any security-testing work inside a Mobix engagement: reviewing captured traffic, planning static or dynamic tests, hunting for IDOR/BOLA, checking JWT/session/token handling, assessing local storage or IPC surface, or writing up findings. Triggers automatically inside `mp engage` sessions (interactive or --auto) and for any manual request to analyse a target app in this lab. Mobix-specific: assumes the mp-traffic / mobix MCP tools and the rooted MuMu Android emulator this repo controls, not generic mobile tooling or another lab's device.

6 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use cloud-pentest, agentic-app-audit and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • shuvonsecSkill

    cloud-pentest

    Post-access cloud exploitation for AWS, GCP, and Azure — what to do AFTER you obtain credentials or reach a metadata endpoint. Covers IAM enumeration and privilege escalation (iam:PassRole, CreatePolicyVersion, AssumeRole chains, GCP service-account impersonation, Azure managed-identity abuse), IMDSv1/v2 metadata credential theft, STS token abuse, S3/GCS/Blob bucket takeover and object exfil, Lambda/Cloud Functions env-var secrets, cross-account and cross-service pivoting, and turning leaked keys into demonstrated impact (read PII, assume admin, exfil data). Assumes authorized scope only. Use when you have AWS/GCP/Azure keys, an SSRF hitting 169.254.169.254 / metadata.google.internal, a leaked service-account JSON, or a token, and need to escalate and prove impact. For finding buckets/origins first, use cloud-recon; for the SSRF entry point, see web2-vuln-classes. 中文触发词:云渗透、AWS提权、IAM枚举、元数据凭证、云安全审计

    Also in 4 projects that use this

  • shuvonsecSkill

    agentic-app-audit

    Black-box security audit of a DEPLOYED AI agent (not the MCP server behind it) — tool-call hijacking, cross-session memory poisoning, confused-deputy via connected tools, agent-to-agent IDOR, excessive agency / unconfirmed destructive actions, and privilege compromise where the agent holds broader perms than the user. Use when the target is a live assistant/agent product with tool access (bookings, email, payments, file/RAG, browsing) rather than a raw LLM chat box or an MCP server you can read.

    Also in 3 projects that use this

  • +10 more from shuvonsec
  • Other versions6 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 6

    1. shuvonsec/claude-bug-bounty★ 5.9ksame name
    2. awarexone/Agentic-Bug-Hunter★ 5.9ksame name
    3. Awarexone/Agentic-Bug-Hunter★ 5.9ksame name
    4. hypnguyen1209/offensive-claude★ 401same name
    5. jessefmoore/offensive-claude-code★ 2same name
    6. noahfranklin/gideon★ 2same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repositoryOne other file it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. 0 are common, 4 are distinctive, and 9 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before doing anything else, work out two things: what's already been done for you, and what mode you're running in.

    line 24
  2. Very specific:

    Read the ranked pass that's already done: Call traffic_brief (or read ./BRIEF.md if it's on disk).

    line 27
  3. Very specific:

    If you have target_*, bypass_*, creds_*, and Bash: you're in an

    line 35
  4. Very specific:

    If you only have traffic_status, traffic_sessions, traffic_flows, traffic_flow, traffic_search, traffic_brief, plus Read/Write/ Grep/Skill: …

    line 38
  5. Very specific:

    (interactive only): If static analysis hasn't run yet, target_static (or mp static) gives you exported components, permissions, network …

    line 58
  6. Very specific:

    Every flow already carries authenticated and auth_scheme.

    line 73
  7. Distinctive:

    Look for inconsistent enforcement: the same resource type authenticated on most calls but not on one.

    line 75
  8. Very specific:

    Check alg, expiry window, and whether the payload carries PII the detector didn't already flag.

    line 90
  9. Very specific:

    If the app signs requests with a custom integrity field (a macIn-style MAC), that's covered in the main README under Re-signing custom …

    line 124
  10. Distinctive:

    Look for tokens, passwords, or PII sitting in plaintext where the app should have used the Keystore.

    line 136
  11. Very specific:

    Use bypass_list to see what's active, and cross-check launch.sh's log markers (root checks blocked: N, Certificate unpinning completed, …

    line 159
  12. Very specific:

    Read-only by default: Don't send a modified request to a live target without the operator confirming scope and blast radius first, same as …

    line 184
  13. Distinctive:

    Stay in scope: Only the target package and its declared in-scope hosts.

    line 189

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Mobile app pentest methodology (Mobix)

This is Mobix's own pentest methodology, shipped with the repo so any mp engage session, on any machine, starts smart rather than generic. It is scoped to what this lab actually does: Android only, against the rooted MuMu emulator described in the main README. Don't reach for iOS guidance, jailbreak tooling, or a physical device farm. None of that exists here.

Orient yourself first

Before doing anything else, work out two things: what's already been done for you, and what mode you're running in.

  1. Read the ranked pass that's already done. Call traffic_brief (or read ./BRIEF.md if it's on disk). This ranks every endpoint by risk with reasoning attached and already splits out broken-access-control, IDOR/BOLA, and mass-assignment candidates. Your job is the layer underneath that: cross-flow correlation, business-logic reasoning, and turning candidates into confirmable, evidenced findings. Don't just restate the brief.
  2. Check which mode you're in. Try listing your available tools, or just attempt lab_status.
    • If you have target_*, bypass_*, creds_*, and Bash: you're in an interactive session (or a hand-started one). Everything in this skill applies.
    • If you only have traffic_status, traffic_sessions, traffic_flows, traffic_flow, traffic_search, traffic_brief, plus Read/Write/ Grep/Skill: you're the --auto agent. No Bash, no other MCP tools, no request sending, by design. Only the traffic-analysis parts of this skill apply to you. Sections marked (interactive only) below are not available to you. Don't try to work around that; it's the deliberate safety boundary for unattended runs. Write up what you can determine from traffic alone, and note in the finding what a follow-up interactive session would need to confirm.

Workflow

Work through these in order. Skip a step outright if nothing in scope touches it, don't force a finding.

1. Map the attack surface

  • traffic_flows grouped by host and path template (or the Traffic tab's site-map tree, or surface.md on disk) to see what's actually been exercised.
  • (interactive only) If static analysis hasn't run yet, target_static (or mp static) gives you exported components, permissions, network security config, and pinning call sites from the decompiled APK. Read AndroidManifest.xml (via jadx output) for exported activities/services/ receivers/providers and custom URL schemes: that's your IPC and deep-link surface, tested in step 7.
  • Note what's missing: endpoints the app's UI clearly supports but that never appeared in traffic mean you haven't exercised that flow yet, not that it's safe.

2. Authorization and access control

This is usually the highest-value category and the most of it is visible from traffic alone.

  • Every flow already carries authenticated and auth_scheme. An endpoint returning 2xx with neither set is the fastest real finding in this lab.
  • Look for inconsistent enforcement: the same resource type authenticated on most calls but not on one. That's a stronger signal than a uniformly open endpoint. It usually means one code path forgot the check.
  • idor_candidates on a flow means the capture feed folded a numeric or hex ID out of the path. Cross-reference across flows: did you see the same endpoint shape hit with more than one ID? If a JWT's sub/role claim doesn't match the ID in the path or body, that's vertical or horizontal privilege escalation waiting to be confirmed.
  • See references/authz-and-idor.md for the full checklist, including what counts as confirmed-from-traffic versus a candidate that needs an interactive Repeater test.

3. Session and token handling

  • Decode every JWT you see (the analysis feed already does this inline). Check alg, expiry window, and whether the payload carries PII the detector didn't already flag.
  • Set-Cookie headers are visible in traffic_flow response headers. Missing Secure, HttpOnly, or SameSite is confirmable directly from traffic; no active test needed for that part.
  • (interactive only) Whether a token survives logout, whether storage on device is Keystore-backed vs. a flat file, and whether a tampered/expired token is actually rejected all need adb/Repeater confirmation.
  • Full detail in references/session-and-tokens.md.

4. Data exposure in transit

  • flags.jsonl already flags JWTs, API keys, PII, and card numbers. Your job is correlating: was the flow authenticated? In scope? Did the endpoint actually need that field for its evident purpose, or is the response over-fetching (a public listing endpoint returning full user objects with emails, for instance)?
  • Access-Control-Allow-Origin: * combined with Access-Control-Allow-Credentials: true on the same response is a directly confirmable CORS misconfiguration; check response headers.
  • Any flow with scheme: http to an in-scope host is cleartext by definition, confirmed, not a candidate.
  • Details in references/data-exposure.md.

5. Business logic and mass assignment

  • Compare request bodies against what the corresponding responses ever return. A field the client sends but no response ever echoes back, especially one named like role, isAdmin, verified, balance, price, or discount, is a mass-assignment candidate.
  • Money- or quantity-bearing endpoints (checkout, transfer, order) where the client supplies the amount are worth flagging even without an active test: note it as a candidate and name the exact Repeater test that would confirm the server doesn't recompute it server-side.
  • If the app signs requests with a custom integrity field (a macIn-style MAC), that's covered in the main README under Re-signing custom request-integrity fields; don't duplicate that here, just use it.
  • Details, including rate-limiting signals, in references/business-logic.md.

6. Local storage (interactive only)

Needs Bash. Not available to --auto.

  • adb shell "run-as <pkg> ls -la /data/data/<pkg>/shared_prefs" and the equivalent for databases/, exactly as documented in the main README's runtime-exploration section.
  • Look for tokens, passwords, or PII sitting in plaintext where the app should have used the Keystore.
  • Full checklist in references/storage-and-platform.md.

7. Platform surface: IPC and deep links (interactive only)

Needs Bash and the exported-components list from step 1's static analysis.

  • Exported activities/services/receivers/providers: try driving them directly with adb shell am start -n <pkg>/<component> (crafted extras) or am broadcast, and watch traffic_flows or logcat for effect.
  • Custom URL schemes: adb shell am start -a android.intent.action.VIEW -d 'scheme://...'. Watch for a deep link that trusts a locally-supplied parameter it shouldn't (an auth token, a user ID, a "skip login" flag).
  • setJavaScriptEnabled(true) combined with addJavascriptInterface() in decompiled source is a WebView JS-bridge red flag worth a targeted look.
  • Full checklist in references/storage-and-platform.md.

8. Client-side protections (interactive only)

This is about assessing the app's own defenses as a finding, not just getting past them to test something else.

  • Use bypass_list to see what's active, and cross-check launch.sh's log markers (root checks blocked: N, Certificate unpinning completed, Proxy configuration overridden) to confirm each one actually engaged for this app, not just that it's toggled on.
  • A high-sensitivity app (banking, health, government-ID) with no meaningful root detection, no tamper/signature check, and no anti-Frida measures is itself a MASVS-RESILIENCE finding worth writing up, even though you can already bypass everything from this lab.
  • Details in references/client-protections.md.

9. Chain and rank

A Low finding that lets you set up a High is worth writing as a chain, not two separate reports. Note the sequence explicitly: what step 1 gives you that makes step 2 possible.

10. Write it up

One file per issue in ./findings/<slug>.md. Format and severity rubric are in references/finding-format.md. Every finding cites evidence by flow number; if you can't point to a specific flow (or a specific static-analysis line), it isn't ready to write up yet, keep investigating or drop it.

Rules

  • Read-only by default. Don't send a modified request to a live target without the operator confirming scope and blast radius first, same as replay.py's dry-run default and the Repeater tab's explicit Send.
  • The --auto agent cannot send requests, period. That's enforced by its tool allowlist, not just a convention; don't try to route around it.
  • Stay in scope. Only the target package and its declared in-scope hosts. Device-wide capture may include other apps' background traffic; that's noise, not signal.
  • Don't exfiltrate real PII into a finding. Reference it by flow number, redacted, the same way the main README's scope-discipline section already asks.

Mobix tool quick reference

NeedTool
Session state, target, scopetraffic_status / lab_status
List / filter captured requeststraffic_flows
One request in full, decodedtraffic_flow
Search bodies/headerstraffic_search
Ranked risk passtraffic_brief
Static analysis (manifest, pinning, exported components)target_static
Toggle/verify a bypass modulebypass_list / bypass_set
Past sessions and engagementslab_sessions
Stored test creds (never returns plaintext)creds_list / creds_send

Full tool list and what each one does: main README, Wiring traffic into your AI tool.

References

  • references/authz-and-idor.md: broken access control and IDOR/BOLA in depth
  • references/session-and-tokens.md: JWT, cookies, session lifecycle
  • references/data-exposure.md: sensitive data in transit, CORS, cleartext
  • references/business-logic.md: mass assignment, tampering, rate limits
  • references/storage-and-platform.md: local storage, IPC, deep links (interactive only)
  • references/client-protections.md: assessing the app's own defenses (interactive only)
  • references/finding-format.md: the exact write-up template and severity rubric

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.