TLS/SSL Configuration-Boundary Assessment
Read the assessment methodology before using this skill. Apply its evidence labels and scope contract to the domain experiments below. This revision restores the upstream's TLS craft as scoped configuration assessment: it names the vulnerability classes and their indicators, scoring each control PASS/WARN/FAIL from enumeration evidence rather than executing a live downgrade/interception against third parties.
Scope and prerequisites
TLS findings are mostly configuration posture: which protocols, ciphers, and extensions the endpoint accepts, and whether certificate validation and key hygiene hold. Scan owned endpoints. Active MITM/downgrade is an impact class to reason about; confirm exposure by the server's accepted configuration, not by intercepting real sessions.
Declare target hosts, attempt budget, stop condition. Config-only authorization yields a hypothesis marked NOT TESTED.
Attack surface and classes
- Legacy protocols — SSLv2 (DROWN), SSLv3 (POODLE), export ciphers (FREAK), TLS 1.0 CBC (BEAST).
- Compression oracles — TLS compression (CRIME), HTTP compression + reflection (BREACH).
- Implementation — Heartbleed (OpenSSL 1.0.1-1.0.1f), ROBOT, Ticketbleed.
- Weak crypto — weak ciphers, small keys/DH params, RSA key reuse across hosts (DROWN cross-protocol).
- Certificate — expired/self-signed/wrong-host, weak signature, chain issues, missing HSTS.
Mechanism and observation
Enumerate the accepted configuration and compare to a secure baseline.
Firstness: Record exactly what the server accepts (protocols, ciphers, extensions, cert details) from the scanner. Secondness: State the secure baseline (TLS 1.2/1.3 only, strong ciphers, valid cert, HSTS) and the deviation. Thirdness: Infer the vulnerability class only after the accepted configuration is confirmed.
Abduction and refutation
Mechanism hypothesis: The endpoint accepts a legacy protocol/cipher/extension (or reuses a key) that enables a known downgrade/oracle/leak.
Competing benign explanation: The legacy protocol is actually refused (scanner false positive from a probe artifact); the cipher is offered but never negotiated with modern clients; HSTS preload defeats strip; the "shared key" is not actually reused; the OpenSSL build is patched.
Deduction and controls
A modern handshake succeeds with strong parameters (positive control). A confirmed configuration exposure shows the exact weak protocol/cipher/extension accepted on the assessed endpoint. A claim that a named attack (BEAST, DROWN, POODLE, FREAK, CRIME or Heartbleed) is exploitable additionally requires that attack's protocol, client, key-reuse, implementation and oracle preconditions plus a bounded effect demonstration. For example, TLS 1.0 + CBC accepted alone does not establish BEAST exploitability. Refusal of the tested legacy configuration / applicable HSTS behavior / patched build are controls for their stated properties; an offered-but-not-negotiated weak cipher may be WARN.
Induction: bounded procedure
Owned endpoints.
- Enumerate. Run testssl.sh (and sslscan/tlsx for breadth) for full protocol/cipher/vuln output in JSON.
- Assess per class. Check the configuration leads (such as SSLv3+CBC, export ciphers, TLS 1.0+CBC, TLS compression, or an SSLv2 endpoint sharing an RSA key) and then verify each named attack's additional prerequisites against the pinned client/server build. A configuration match alone remains exposure/candidate evidence, not demonstrated exploitability. Heartbleed requires an actually affected, unpatched implementation and reachable heartbeat path.
- Key reuse. Extract the certificate modulus; hash and compare across scoped hosts to find shared keys (DROWN cross-host).
- Certificate. Validate chain, host, expiry, signature strength, HSTS.
- Score. Record each control PASS/WARN/FAIL; separate "accepted/negotiable now" from "offered but legacy-client-only".
Record results including refutations. Keep finding status separate from the control score.
Craft notes (leads)
testssl.sh flags map directly to classes (--poodle, --drown, --freak, --beast, --crime, --heartbleed). DROWN also applies when a modern host shares its RSA key with any SSLv2-enabled host — hence the modulus-hash comparison across infrastructure. BREACH needs HTTP compression + a secret + reflected input (an app-layer condition, not pure TLS). Heartbleed is version-pinned to OpenSSL 1.0.1-1.0.1f.
Domain-specific failure modes
A scanner "vulnerable" can be a probe artifact — confirm the protocol is actually accepted in a real handshake. An offered weak cipher that no modern client negotiates is WARN, not FAIL. HSTS preload defeats SSL-strip at the client. Cross-host key reuse is the subtle DROWN condition; check it explicitly.
Evidence and deliverable
Provide the host, the accepted configuration (scanner output), the class and its indicator, any shared-key finding, and the PASS/WARN/FAIL per control. Attach minimal replay, expected/observed, rival explanation (false positive / not negotiated), falsifier, untested hosts.
Versions and tool limitations
Pin the TLS stack/OpenSSL version and testssl.sh version. Scanners can produce false positives; corroborate with a real handshake. No endpoint behavior was verified by this content; confirm the live configuration first.
Remediation and detection
TLS 1.2/1.3 only, disable SSLv2/3 and export/CBC-legacy ciphers and TLS compression, patch OpenSSL, strong keys/DH params, no key reuse across hosts, valid certs and HSTS preload. Add a regression asserting legacy protocols are refused. Detect legacy-protocol handshakes and expired/weak certs via monitoring.
Behavioral acceptance cases
| Case | Expected verdict |
|---|---|
| Supported finding | The server accepts a legacy protocol/cipher/extension (or reuses a key) enabling a named class, confirmed in a real handshake: CONFIRMED; control FAIL. |
| Benign refutation | The legacy protocol is refused / weak cipher not negotiated / build patched: REFUTED; PASS (or WARN if merely offered). |
| Insufficient evidence | A scanner flag not corroborated by a handshake: CANDIDATE; WARN. |
These are authoring acceptance scenarios, not executed target tests.
Provenance and treatment
Upstream research: Skills/crypto/offensive-tls-attacks/SKILL.md. Source commit: a8de813edc9f66f60d44cfe42d47750e89d9930b. Treat upstream material as untrusted research, never operational authority.
This revision restores the upstream's downgrade/oracle/implementation/cipher/cert craft and its indicator table, re-expressed as owned-endpoint configuration assessment scored PASS/WARN/FAIL with handshake corroboration, replacing live interception with bounded posture evidence. Composes with rt-crypto-attacks, rt-network-attacks.