К основному содержанию
скилл CX Codex

mid-engagement-ir-detection — скилл для Claude Code: IR на пентесте

Скилл mid-engagement-ir-detection для Claude, который учит замечать, как цель меняется по ходу теста: патчи SOC, новые правила WAF, чужой атакующий рядом — и превращать это в находки.

MIT ★ 16 193 обновлён 2026-10-29

Установить в Claude Code · Через маркетплейс плагинов Все способы и платформы ↓
/plugin marketplace add elementalsouls/Claude-BugHunter
/plugin install claude-bughunter@elementalsouls
Открыть на GitHub

Что делает mid-engagement-ir-detection

Скилл описывает, как следить за самой целью во время активного теста. Защищённое приложение меняется на ходу: SOC ставит патч, включается новое правило WAF, кто-то посторонний в это же время перебирает учётки. Такие сдвиги скилл предлагает оформлять как отдельные находки, а не списывать на ложное срабатывание.

Устроено просто. До первого запроса снимается отпечаток цели (время и размер ответа, заголовки, cookie, счётчик блокировок) и кладётся в engagement_log/fingerprint_pre.json. Во время теста каждый результат дописывается в JSONL, старые записи не правятся. После сессии или на первой неудачной перепроверке считается дельта по времени, размеру, новым заголовкам и новым блокировкам. Значимая дельта — повод разбираться, а не отзывать находку.

Для подозрения на правило WAF скилл даёт список обходных вариантов: другая кодировка payload, смена метода, multipart вместо form-urlencoded, пауза между запросами, регистр ключевых слов. Если сигнал вернулся, защита стоит на уровне WAF и код по-прежнему уязвим. Если нет, правка, скорее всего, в коде. Скилл рассчитан на авторизованную работу, где клиент знает о тесте: в баг-баунти сравнивать состояние до и после не с чем.

Какие задачи решает

  • заметить, что подтверждённая SQLi перестала воспроизводиться, и не отозвать находку
  • отличить правило WAF от правки в коде по поведению payload
  • посчитать новые блокировки учёток за сессию и понять, что рядом работает второй атакующий
  • снять отпечаток цели до теста, чтобы потом было с чем сравнивать
  • оформить реакцию SOC отдельной находкой с оценкой глубины защиты

Примеры запросов агенту

  • Сними отпечаток цели перед началом активного теста и сохрани его в engagement_log/fingerprint_pre.json.
  • Подтверждённая SQLi перестала воспроизводиться на перепроверке — разберись, это правило WAF или правка в коде.
  • Посчитай новые AADSTS50053 за сессию и скажи, мог ли я сам вызвать Smart Lockout.

Как установить mid-engagement-ir-detection

1Через маркетплейс плагинов

/plugin marketplace add elementalsouls/Claude-BugHunter
/plugin install claude-bughunter@elementalsouls

Команда ставит плагин claude-bughunter целиком, со всеми его скиллами

2Через Skills CLI (Vercel)

npx skills add elementalsouls/Claude-BugHunter --skill mid-engagement-ir-detection -a claude-code

3Вручную: положить файлы

~/.claude/skills/mid-engagement-ir-detection/SKILL.md

Папка скилла целиком; для одного проекта — .claude/skills/mid-engagement-ir-detection/

По шагам, с командами для каждой платформы: Как установить скилл в Claude Code: пять способов

Вопросы о mid-engagement-ir-detection

Как установить mid-engagement-ir-detection в Claude Code?

Через маркетплейс: сначала /plugin marketplace add elementalsouls/Claude-BugHunter, затем /plugin install claude-bughunter@elementalsouls.

Что делает mid-engagement-ir-detection?

Учит фиксировать изменения состояния цели во время активного теста и превращать их в находки: патч SOC, свежее правило WAF, блокировки чужих учёток.

Как понять, что блокировки учёток вызвал не я?

Скилл держит одну попытку на пользователя и сверяется с логом инструмента, что дублей не было. Smart Lockout срабатывает после 10 неудачных попыток, поэтому новые AADSTS50053, накопившиеся за вашу сессию, пришли от кого-то другого.

Исходный текст

SKILL.md · MIT
---
name: mid-engagement-ir-detection
description: Methodology for detecting client SOC patches, attacker activity, and security-state changes that occur DURING a red-team engagement — and converting those observations into deliverable findings. Built from authorized red-team work where the client patched a confirmed SQLi within 30 minutes of detection AND an external attacker locked multiple new accounts during a single test session. Use when (a) running ANY active engagement against a monitored target, (b) a previously-confirmed finding stops reproducing, (c) baseline timing shifts unexpectedly, or (d) you notice response patterns changing during testing.
sources: authorized-engagement
report_count: 1
---

## When to use this skill

Trigger when:
- Running active testing against a target with active SOC monitoring
- A confirmed-vulnerable finding stops reproducing on recheck
- Baseline timing shifts unexpectedly (3× slower, sudden errors, new headers)
- Response sizes change between test windows
- New WAF cookies or headers appear that weren't there at session start
- Lockout / error rates change between test windows (especially LOCKED count for credential attacks)
- Engagement is "assume breach" or "white box" — client knows you're testing

DO NOT use for:
- Bug bounty (client doesn't know you're there; no real-time IR)
- Pure recon (no state-change happening)
- One-off vulnerability scanning (no temporal dimension)

---

## The core insight

In a real red-team engagement against a competent SOC, the security state of the target is **not static**. It changes during your test in response to your traffic. These state changes are:

1. **Themselves valuable findings** (positive operational observations about IR responsiveness)
2. **Confirmation evidence** (mid-engagement patch = the original vulnerability was real)
3. **Classification signals** (WAF rule deployment vs code fix — different remediation depth)

Anti-pattern: treating reproduction failure as evidence the original signal was a false positive. **Original PoC artifacts captured before the change are still the vulnerability finding.**

---

## The discipline — capture before, diff after

### Before any active test:

```python
# Capture pre-test fingerprint of the target
fingerprint = {
    "ts_pre": time.time(),
    "ip_seen": "<operator-src-ip>",
    "baseline_response_time_ms": <measure>,
    "baseline_response_size_bytes": <measure>,
    "response_headers": <capture set>,
    "waf_cookies": <list>,
    "lockout_count_in_state": <count from o365_attempts.json>,
}
```

Persist to `engagement_log/fingerprint_pre.json`.

### During the test:

Log every test result with full context (timestamp, IP, payload, response code, response size, response time, headers if relevant) to JSONL append-only.

### After the test session OR on first failed-recheck:

```python
fingerprint_post = same structure
delta = {
    "baseline_time_change_ms": post.time - pre.time,
    "baseline_size_change_bytes": post.size - pre.size,
    "new_headers_appeared": post.headers - pre.headers,
    "new_waf_cookies": post.cookies - pre.cookies,
    "new_lockouts": post.locked_count - pre.locked_count,
}
```

If any delta is significant — **investigate, don't retract**.

---

## The three primary IR observations

### Observation 1 — Mid-engagement WAF rule deployment

**Symptoms:**
- Original payloads return identical response → no signal at all on recheck
- Body size identical to baseline (login page reflection)
- Timing reverts to baseline regardless of payload
- New cookie or header in responses (e.g., `cf-bm`, `__cf_bm`, `awselb`)
- Specific keyword in URL/body now triggers different response code (403, 406, 429)

**Confirmation:** retry with WAF-evasion variants:
- URL-encode the payload differently (`%27` vs `%5cu0027`)
- Change request method (POST → PUT, GET → POST)
- Different content-type (form-urlencoded → multipart)
- Slower pace (5s → 60s between requests)
- Mixed-case keywords (`SLEEP` → `SlEeP`)

If WAF-evasion variants restore the signal, the mitigation is at the WAF layer (bypassable).

If even WAF-evasion variants stay blocked, the mitigation is likely in code.

**Finding template:**
```
Subject: Mid-engagement mitigation deployed for <vulnerability X>
Observation: At engagement timestamp T0, vulnerability <X> on <endpoint> was
confirmed via <PoC>. At T0+<minutes>, recheck via the original payload no longer
reproduces the timing/error/size differential. <WAF-evasion variant> [does/does
not] restore the signal.

Description: This pattern is consistent with the client SOC observing engagement
traffic and deploying a mitigation in real time. Mitigation depth assessment:
[at-WAF, bypassable] vs [in-code, durable].

Impact (positive): Client SOC has both detection-grade visibility into application
traffic AND the authority to deploy mitigations within ~<minutes> of detection.

Impact (caveat): The original vulnerability did exist and was exploitable for at
least the engagement window before mitigation. If the mitigation is at the WAF
layer only, the underlying code-level flaw remains exploitable via alternative
payloads.

Recommendation:
1. Verify the mitigation is in code (parameterized queries, input sanitization),
   not just at the WAF layer.
2. Audit the codebase for the same root cause across sister applications.
3. (Positive) Document the IR responsiveness as a capability metric.
```

### Observation 2 — Active concurrent attacker

**Symptoms:**
- Many `AADSTS50053` (LOCKED) responses despite your 1-attempt-per-user discipline
- Lockouts cluster alphabetically or by some other sort key
- New lockouts appear DURING your engagement (diff before/after)
- LOCKED rate exceeds expected baseline (in our engagement: 11% of all attempts → red flag)

**Math check:**
- Your discipline: 1 attempt per user lifetime
- Smart Lockout default: lockout after 10 failed attempts; lockout duration starts at 60 seconds and grows with each subsequent lockout (not a flat 10-minute window)
- Therefore: IF tool logs confirm exactly 1 attempt/user (no burst retries, no parallel-goroutine duplicate sends, no tool misconfiguration), you cannot mathematically cause Smart Lockout — verify `journal.jsonl` shows 1 attempt/user before asserting this
- Therefore: every NEW AADSTS50053 accumulating during the window (per before/after diff) was caused by someone else; pre-existing locks may stem from the legitimate user's own failures or a prior test run, so attribute only the newly-accumulating locks to an external party

**Confirmation:**
- Sort locked accounts alphabetically; if they cluster, attacker is using sorted username list
- Compare pre-session lockout count vs post-session — new locks during your session = attacker is active *right now*
- Probe a known-active "system" account (`noreply@`, `info@`, `oc@`) — if it's locked, attacker is hitting service mailboxes too (typically MFA-exempt → high-value to attackers)

**Finding template:**
```
Subject: Active external password-spray campaign detected during engagement

Observation: During M365 ROPC validation against the <tenant> Entra tenant, <N>
unique principals returned AADSTS50053 (Smart Lockout) when probed with a single
password attempt at safe pace. Our tool journal (journal.jsonl) confirms exactly
1 attempt per user with no duplicate sends, so under the Smart Lockout default
(lockout after 10 failed attempts) we cannot have caused these lockouts. The
attribution below applies to the <K> NEW locks that accumulated during our
engagement window between <timestamp_start> and <timestamp_end> (per before/after
diff); pre-existing locks are not attributed to this campaign.

Description: The pattern (alphabetical clustering, real-time accumulation,
including system mailboxes) is consistent with an external attacker performing
a username-list-driven password spray attack against the tenant.

Impact: An external adversary is actively attempting to compromise corporate
M365 accounts. The attacker has knowledge of the user-email schema and a
password-guess wordlist. <List of locked accounts is now in attacker's hands>
(Smart Lockout differentiates valid from invalid usernames).

Recommendation (CRITICAL — within 24h):
1. Open a P1 incident with the SOC. Pull Entra sign-in logs for the <N> locked
   accounts over the last 30-60 days. Identify source IPs and time windows.
2. Apply Conditional Access rule blocking sign-ins from outside <expected geo>
   for non-admin accounts.
3. Enable Identity Protection's User Risk policy with auto-reset on high risk.
4. Force tenant-wide password reset for all <N> previously-locked accounts.
5. Audit service accounts for MFA exemptions; ensure all human-interactive
   accounts have phishing-resistant MFA.

Evidence: engagement_log/poc/m365/locked_accounts.txt (<N> entries with timestamps)
```

### Observation 3 — Detection-induced rate limiting / IP blocks

**Symptoms:**
- Specific IP starts returning 403 / 429 / 451 after a window of normal responses
- Specific IP starts seeing dramatically slower responses (3x+ baseline)
- TLS handshake fails or RST mid-connection
- DNS suddenly returns NXDOMAIN for hosts that resolved before
- Some hosts work from one IP but not another

**Confirmation:**
- Rotate to a different IP and retry — if works, you got rate-limited/blocked
- Compare TTL on DNS responses — sudden short TTL = active mitigation deployed
- Check `Server:`, `Via:`, `CF-Cache-Status:` headers for CDN-introduced limits

**Finding template (operational note, usually low/info severity):**
```
Subject: Engagement traffic detected — IP <X> rate-limited at <timestamp>

Observation: After ~<N> requests in <window> from IP <X>, target <hostname>
began returning <code> for all subsequent requests. Rotation to IP <Y>
restored normal responses.

Description: Active anti-automation control at the perimeter (CDN/WAF/origin).

Impact (positive): Volumetric anti-automation is functional.

Impact (caveat): Rotation defeats this control trivially (cloud VMs cost <$5).
A patient adversary or spray-from-residential-proxies attacker is not affected.
```

---

## State diff — the key technique

Maintain three pieces of state:

### 1. `engagement_log/baseline.json` — captured at session start

```json
{
  "ts": "2026-05-08T13:00:00",
  "source_ip": "<operator-src-ip>",
  "targets": {
    "https://target.example.com/login": {
      "baseline_response_time_ms": 584,
      "baseline_response_size_bytes": 11966,
      "headers_seen": ["Server: Apache", "X-Powered-By: PHP/8.0.26"],
      "set_cookie_names": ["PHPSESSID"]
    }
  },
  "m365": {
    "lockout_count": 247,
    "valid_creds_count": 0
  }
}
```

### 2. `engagement_log/journal.jsonl` — append-only test log

Every test logs:
```jsonl
{"ts":"...","ip":"...","tool":"...","target":"...","payload":"...","resp_code":...,"resp_size":...,"resp_ms":...,"verdict":"...","notes":"..."}
```

### 3. `engagement_log/state_changes.jsonl` — observed deltas

When a state change is detected, append:
```jsonl
{"ts_observed":"2026-05-08T14:30:00","change_type":"baseline_time_shift","target":"https://<employee-app-host>/<app>/login.php","baseline_pre_ms":24412,"baseline_post_ms":90403,"interpretation":"likely WAF rule deployed","actions_taken":["tested WAF-evasion variants — no signal restoration","documented as IR-mitigation finding"]}
```

This third file is your finding evidence. Every entry is a candidate finding.

---

## Tooling — automated state-change detection

```bash
# Bash watcher — runs every 5 min during engagement, alerts on shifts

cd "$ENGAGEMENT_DIR"

# Re-measure baseline timing on key targets
for target in "$@"; do
  ms=$(curl -sk -o /dev/null -w "%{time_total}" "$target" --max-time 30)
  ms_int=$(echo "$ms * 1000" | bc | cut -d. -f1)
  echo "$(date -u +%FT%TZ) $target $ms_int" >> baseline_history.log
done

# Diff against pre-session baseline
python3 - << 'PY'
import json, time
baseline = json.load(open("engagement_log/baseline.json"))
for line in open("baseline_history.log"):
    parts = line.strip().split()
    ts, target, ms = parts[0], parts[1], int(parts[2])
    if target in baseline["targets"]:
        pre = baseline["targets"][target]["baseline_response_time_ms"]
        if abs(ms - pre) > pre * 0.5:  # >50% shift
            print(f"ALERT {ts} {target} time {pre} -> {ms}")
PY
```

For lockout-count tracking on M365:
```bash
# Run every 30 min during M365 spray
LOCK_COUNT=$(grep -c '"AADSTS50053"' engagement_log/o365_results.jsonl)
echo "$(date -u +%FT%TZ) lockout_count $LOCK_COUNT" >> lockout_history.log
# Diff first vs last to surface delta
```

---

## The "single signal recanted" rule

If a confirmed-vulnerable finding stops reproducing:

1. **DO NOT delete the original PoC.** Original timestamps + payloads + response captures are forever.
2. **Capture the new state in detail.** What's the recheck response? What's different?
3. **Try at least 3 alternative vectors** before declaring "indeterminate".
4. **If none restore the signal**, document as "vulnerability confirmed at T0, mitigation observed at T0+<delta>, mitigation depth: [WAF | code]".
5. **Both states go in the report.** The original finding + the IR observation.

This is the discipline that distinguishes professional red team from "hobbyist scanning". Your client wants the timeline of vulnerability + mitigation, not just the static state.

---

## Why this is a finding (selling it to the client)

When you tell a client "I confirmed SQLi at 14:24, you deployed a mitigation by 14:55, here's the original PoC and here's why you should still verify the fix is in code":

- **Confirmed the vulnerability existed** (auditor-grade evidence)
- **Confirmed the mitigation was deployed** (positive ops finding for the SOC)
- **Identified the depth question** (WAF vs code — different remediation cost)
- **Demonstrated red-team value** (you proved both the bug AND the IR responsiveness)

This is a more valuable deliverable than "I confirmed SQLi" alone, because it captures the engagement's full operational picture.

---

## Anti-patterns to avoid

- **"Recheck failed → false positive."** No. Recheck failed because the target changed. Investigate the change.
- **"It's no longer vulnerable, drop the finding."** No. It WAS vulnerable. The finding stays. Add a "current state: mitigated" annotation.
- **"They patched it, mission accomplished."** No. WAF rule != code fix. Verify the depth.
- **"Don't tell the client we observed their patch."** Yes, do. It's a positive finding about their IR.
- **"Don't include the locked accounts list — they'll think we caused it."** No. Math + journaling discipline proves you didn't. Include the list with the math.

---

## Bridge to neighboring skills

- `redteam-mindset` — broader discipline framework; this skill is a specific application
- `m365-entra-attack` — specific case for M365 / Smart Lockout differential
- `evidence-hygiene` — how to capture and redact PoC evidence properly
- `report-writing` — finding template for IR observations
- `bb-methodology` — note that this skill is INAPPROPRIATE for bug bounty (no real-time IR there)

---

## One-line summary

**Your engagement leaves a footprint. The footprint changes the target. Capture both states. Both are findings.**

---

## Related Skills & Chains

- **`redteam-mindset`** — This skill is a specific application of the broader red-team discipline. Engagement flow: `redteam-mindset` loaded at engagement start → baseline-capture habit built in → when response patterns shift mid-test, `mid-engagement-ir-detection` activates to capture the SOC-patch state as a NEW finding (defensive-action observed = client capability metric, not "the bug got fixed so we lose the finding").
- **`evidence-hygiene`** — Mid-engagement IR detection produces TWO states (pre-patch and post-patch); both need disciplined evidence capture or the second finding can't be defended. Engagement flow: baseline screenshots + timestamped request/response dumps at session start → response shift detected → second capture set with explicit timestamp delta → both packaged together.
- **`m365-entra-attack`** — The single richest source of mid-engagement IR signal in modern engagements. Engagement flow: M365 spray triggers AADSTS50053 lockout → baseline lockout policy captured → if lockout window changes mid-test (e.g., from 60min to 24hr) → `mid-engagement-ir-detection` captures the policy change as a finding ("CA policy hardened mid-engagement; defensive response measured").
- **`enterprise-vpn-attack`** + **`vmware-vcenter-attack`** — Critical-infrastructure CVE exploitation is the highest-noise activity; expect SOC to patch within hours. Engagement flow: confirmed VPN/vCenter CVE → baseline capture BEFORE exploitation attempt → if appliance updates mid-test, capture as defensive-action finding → report both the original CVE AND the IR-response.
- **`redteam-report-template`** — IR-observation findings get their own Subject in the deliverable, framed differently from technical-vuln findings. Engagement flow: `mid-engagement-ir-detection` captures behavior-change event → `triage-validation` 7-Question Gate (specifically: "is the behavior-change attributable to my activity?") → `redteam-report-template` packages as a "client capability observation" with explicit timeline and detection-latency metric.

© Sachin Sharma · MIT · elementalsouls/Claude-BugHunter

Похожие скиллы

Все скиллы по теме «Безопасность» →
exploit-guide 0xSteph Субагент exploit-guide консультирует по методикам эксплуатации уязвимостей в тестах на проникновение с обязательным разбором индикаторов компрометации. CX Codex ★ 2 482звёзд репо engagement-planner 0xSteph Субагент engagement-planner планирует авторизованный пентест по фазам, привязывает техники к MITRE ATT&CK и составляет шаблон правил проведения работ (RoE). CX Codex ★ 2 482звёзд репо red-team-engagement briiirussell Скилл red-team-engagement помогает спланировать и провести авторизованную симуляцию атак red team для проверки защиты. Пригодится при составлении правил engagement, сценариев assumed-breach и деконфликтации. CX Codex +1 439установок conducting-full-scope-red-team-engagement mahipal Скилл conducting-full-scope-red-team-engagement для Claude Code планирует red team операции по MITRE ATT&CK. Запускается для оценки защиты от APT-атак или эмуляции действий группировки. CX Codex +1 243установок security-case-management elastic Скилл security-case-management управляет инцидентами в Elastic Security через Kibana Cases API. Скилл создаёт кейсы, ищет связанные алерты и добавляет заметки расследования прямо в чате агента. CX Codex +1 4 600установок incident-response Alireza Rezvani Скилл incident-response классифицирует сработавшую тревогу по 14 типам инцидентов, отсеивает ложные и подсказывает, кого звать и что фиксировать. CX Codex +1 634установок

Из того же набора92

Весь набор →
hunt-lfi Скилл hunt-lfi помогает искать LFI, RFI и обход путей, проверяя, что доступ к файлу действительно сработал. Используйте его при аудите уязвимостей включения файлов. CX Codex +1 ★ 16 193звёзд репо hunt-dom Скилл hunt-dom для поиска клиентских DOM-уязвимостей: DOM Clobbering, PostMessage без проверки origin, CSS Injection с утечкой токенов. CX Codex +1 ★ 16 193звёзд репо enterprise-vpn-attack Скилл enterprise-vpn-attack пентеста SSL VPN-шлюзов: Cisco ASA, Fortinet, Citrix NetScaler, Palo Alto GlobalProtect, Pulse Secure/Ivanti и другие — от распознавания вендора до CVE пред-авторизации. CX Codex +1 ★ 16 193звёзд репо hunt-grpc Скилл hunt-grpc помогает проверять gRPC-сервисы на проблемы с отражением, авторизацией и доверием к прокси. Пригодится при аудите микросервисов с HTTP/2. CX Codex +1 ★ 16 193звёзд репо hunt-business-logic Скилл hunt-business-logic помогает находить уязвимости бизнес-логики в веб-приложениях и платёжных шлюзах. Пригодится при аудите чекаутов и систем верификации. CX Codex +1 ★ 16 193звёзд репо hunt-dispatch Скилл hunt-dispatch для оркестратора /hunt: выбирает навыки проверки по сигналам с целевого хоста. Полезен при запуске сценария Red Team или WAPT. CX Codex +1 ★ 16 193звёзд репо apk-redteam-pipeline Скилл apk-redteam-pipeline автоматизирует реверс Android-приложений: скачивает APK с Play Store, apkpure и apkmirror, декомпилирует через jadx, ищет секреты, JWT, Firebase-ключи, экспортированные компоненты и сертификаты. CX Codex +1 ★ 16 193звёзд репо hunt-brute-force Скилл hunt-brute-force проверяет защиту эндпоинтов от перебора паролей, OTP и перечисления пользователей. Помогает найти отсутствие лимитов запросов и выявить скрытый троттлинг. CX Codex +1 ★ 16 193звёзд репо

Спонсоры проекта

Компании, которые помогают держать аналитику открытой