ios-security-research

Autonomous iOS security research workflow. Triggers on iOS bug-bounty, mobile pentest, app reverse-engineering, API testing, IDOR, auth bypass, mass assignment, GraphQL probing, JWT analysis, runtime

SKILL.md

GitHub

iOS Security Research — Orchestrator Skill

You are the orchestrator for an autonomous iOS security engagement. Use this skill when the user asks for iOS app reverse engineering, mobile API testing, bug-bounty work against an iOS target, or any workflow that combines runtime hooking with HTTP traffic analysis.

Source-of-truth files

Read these first when a user starts an engagement. Their contents define how the platform behaves.

  1. docs/architecture.md — module map and design constraints.
  2. docs/workflows.md — the six engagement phases (bootstrap → passive → mapping → active → exploit → report) and what runs in each.
  3. docs/bug-bounty-modules.md — IDOR, auth, mass assignment, tamper, GraphQL, token analysis — what each tests and how findings are emitted.
  4. docs/frida-objection-integration.md — Frida hook library, Objection script library, and when to use each.
  5. docs/mcp-integration.md — the vendored mitmproxy-mcp surface and the iOS-aware additions.
  6. skills/_upstream/anthropic-cybersecurity-skills/skills/reverse-engineering-ios-app-with-frida/SKILL.md — procedural Frida reference.
  7. skills/_upstream/anthropic-cybersecurity-skills/skills/analyzing-ios-app-security-with-objection/SKILL.md — procedural Objection reference.

How to act

When the user describes an engagement, do this:

Step 1 — confirm environment

openrecon doctor

If anything fails: read docs/roadmap.md and follow the setup steps. Do not proceed until the doctor is clean.

Step 2 — start the engagement

openrecon run --target <bundle_id> --device <usb|id> --budget 1800

This calls agent.runner.run_engagement, which:

  1. Spawns the target via Frida (frida_layer.runner.FridaRunner).
  2. Starts the vendored mitmproxy-mcp (mitm.client.MitmClient).
  3. Ingests events through agent.correlate.Correlator into runs/<run_id>/.
  4. Ticks agent.planner.Planner.next_step() until the engagement is terminal.

The planner walks through phases automatically. You don't need to drive it step-by-step — but you do need to watch the output. Each step writes to runs/<run_id>/state.json.

Step 3 — read findings

ls runs/<run_id>/findings/
cat runs/<run_id>/report.md

Summarize the findings for the user. Link to each finding's Markdown file. Group by severity. Quote a representative flow_id per finding so the user can openrecon replay <finding_id> to confirm.

Step 4 — drill in if needed

If the user wants to investigate a specific finding or endpoint:

  • openrecon correlate <run_id> — recompute correlations after tuning.
  • python -m api.idor --run-dir runs/<run_id> --baseline <flow_id> — re-run a module on its own.
  • Open the relevant Frida hook in frida_layer/hooks/<hook>.js and modify it; restart the engagement.

Boundaries

  • Don't write throwaway code in chat. Add hooks to frida_layer/hooks/, modules to api/, addons to mitm/addons/. Then invoke.
  • Don't modify mitm/vendor/ or skills/_upstream/ directly. Patches go into the layer that wraps them (mitm/addons/, mitm/client.py, or our own SKILL.md wrappers).
  • Don't bypass correlation. Findings without a correlated Frida event are weaker. If a module emits one without correlation, that's a signal something is wrong with the run.
  • Don't add a UI. This platform is CLI + Claude Code only.

Common requests and how to map them

User says You do
"Test this iOS app for IDORs" openrecon run with budget, then once mapping is done, python -m api.idor --run-dir runs/<id> --baseline <flow_ids>
"Check the auth on this endpoint" Replay that specific flow with api/auth.py
"Decrypt the request body" Look at commoncrypto_tracer.js events; if HMAC/AES seen, the input hash should match request body sha256
"Bypass jailbreak detection" jailbreak_bypass.js is loaded by default; if app still detects, look at objection ios jailbreak disable for the broader pattern set
"What does this method do" objection ios hooking watch class <X> via objection_layer.runner
"Replay this request as user B" mitm.replay.ios.replay_as_ios(...) with set_headers containing user_b's Authorization

Output expectations

Every engagement produces:

  • runs/<run_id>/frida_events.jsonl
  • runs/<run_id>/mitm_flows.jsonl
  • runs/<run_id>/correlations.jsonl
  • runs/<run_id>/findings.jsonl
  • runs/<run_id>/findings/<id>.{md,json} for each finding
  • runs/<run_id>/report.md + report.json

When summarizing for the user, link to each of these — they are the audit trail.