oracle-verification-technique
Machine-oracle verification for security findings: promote a finding to VERIFIED only when a named oracle re-runs the exploit N/N against the live target AND a negative control fails on a safe surface, then package a portable proof capsule the client can replay. Use before reporting web/API/auth fin
v2026.11.13-21cd5682 installazioni+2 in 7 giorni39 StelleMITsincronizzato 11 ore fa
Installa
Non hai ancora la CLI di NotHoly? Installala prima.
$ curl -fsSL https://notholy.com/install.sh | sh$ notholy skills install aeondave/oracle-verification-techniqueScrive in ~/.notholy/skills dopo aver controllato la firma e lo SHA-256.
FirmaFirmato, ed25519
Pacchettoad066328…ffa33feb
ad0663284f97…9757ffa33feb3 file · 6.4 KBScansioneBasso1 segnalazione
- Link a domini esterni
Sorgentegithub.com/AeonDave/malskill/tree/main/offensive-techniques/oracle-verification-technique@21cd568
Questa voce non è ancora scritta in italiano, quindi è mostrata in un'altra lingua: English.
In questa pagina
Goal: A finding is a candidate until a named machine oracle reproduces it against the live target N-out-of-N times AND a negative control fails on a safe surface. Only then is it VERIFIED, shipped with a proof capsule anyone can replay. An LLM assertion, a scanner hit, or a single lucky exploit is never proof on its own.
When this technique applies
- Before reporting any web/API/auth/injection finding as confirmed.
- Triaging third-party scanner output (nuclei, nikto, zap, sqlmap, dalfox) — hold it back until re-proven.
- An LLM or delegated sub-agent claims a vulnerability without a reproducible receipt.
- A report must split VERIFIED from candidate, or a CI gate must break only on proven findings.
- Multi-step chains where each hop must be independently proven before the chain is claimed.
Do not use this to find bugs — it only decides whether a suspected bug is real. Discovery stays with the web-exploit / vuln-exploit / recon techniques; this gates their output.
Core rule (enforce, don't just intend)
No verdict without a named oracle. A
VERIFIEDbadge that cannot name the oracle that earned it is rejected and downgraded tocandidate. The model coordinates; the oracle decides what is true.
Encode this as a hard check in your workflow and in report generation, not as a hope.
The verification loop
For each candidate finding, run these gates in order. Fail any gate → stay candidate.
1. Name the oracle
Pick a deterministic oracle for the vuln class — a check that returns a machine-decidable positive/negative from the target's actual response, not a pattern match on a scanner label. For an application-response oracle, reject a 200 OK challenge or block attributed to the WAF/CDN rather than the application. A generic access-denied body alone does not establish its origin; compare known application and blocked baselines before counting the signal. If no oracle exists for the class, the finding cannot be VERIFIED (report it as candidate — no oracle). Load references/oracle-catalog.md for per-class oracles and their positive signals.
2. Re-run N/N against the live target
Re-execute the exploit recipe N times (default N=3) against the real target. Every run must produce the positive signal. Any miss → candidate (flaky / not reliably exploitable). Fresh session/token each run where auth is involved — a replay that only works with a stale cached token is not proven.
3. Negative control must fail on a safe surface
Run the same oracle against a surface that must be safe: a non-vulnerable endpoint, a patched route, a benign parameter value, or a control payload that should do nothing. The control must fail (no positive signal). If the oracle fires on the safe surface too, the oracle is broken or the signal is ambient — the finding is not proven. This is what stops false positives; do not skip it.
4. Separate proof from impact
A bare signal (an out-of-band callback, a reflected marker) proves reachability, not harm. Rate reachability-only findings medium until an impact oracle reproduces real impact (data read, auth bypass, RCE). Do not assume impact from a probe.
5. Package the proof capsule
Emit a portable, replayable capsule (see references/proof-capsule.md): oracle name, exact recipe (request/command + payload), positive signal, the N/N run log, the failing control, and a one-command replay. The client/dev replays it against the live target without trusting your tool. Multi-step chains bundle every proven hop into one capsule.
Scope safety during verification
- Host-lock active oracles (sqlmap, dalfox, raw HTTP) to the declared engagement target. Never feed third-party URLs scraped from page content into an attack oracle.
- Re-running exploits is active testing — respect engagement posture (pentest vs red-team vs lab). Confirm before any oracle that is destructive or irreversible.
- If an aggressive discovery sweep knocked the target over, wait for it to answer again before running oracles, or every gate fails and you falsely report zero.
Output contract
Report each finding with an explicit status:
VERIFIED— oracle named, N/N passed, control failed, capsule attached.candidate— suspected but not (yet) proven; state which gate it failed.candidate — no oracle— no oracle exists for this class; manual verification required.
Never let a scanner-labeled or LLM-asserted finding reach the VERIFIED column.
Pair with sibling skills
offensive-techniques/web-exploit-technique/SKILL.md— supplies the exploit recipes this skill gates.offensive-techniques/vuln-exploit-technique/SKILL.md— impact validation for the impact oracle (gate 4).offensive-techniques/report-generation-technique/SKILL.md— consumes the VERIFIED/candidate split and the capsules.- Behavioural gates
evidence-before-claimsandverification-before-completion— this skill is their enforced, machine-checked form for findings.
Resources
references/
references/oracle-catalog.md— per-vuln-class oracles: positive signal, recipe skeleton, and the negative control to run. Load when choosing or building an oracle (gate 1–3).references/proof-capsule.md— the proof-capsule schema and replay contract. Load when packaging a VERIFIED finding (gate 5).
File di questa versione
references/29.2 KB
- oracle-catalog.md5.3 KB
- proof-capsule.md3.8 KB
- SKILL.md6.0 KB
Permessi
| Permesso | Dichiarato | Trovato dalla scansione |
|---|---|---|
| Legge file | No | No |
| Scrive file | No | No |
| Esegue comandi shell | No | No |
| Usa la rete | Sì | Sì |
| Usa i connettori | No | No |
| Controlla un browser | Sì | Sì |
Segnalazioni della scansione
Rivisto il 13 nov 2026
- InfoLink · Link a domini esterni
Una scansione pulita abbassa il rischio, ma non prova che il pacchetto sia sicuro. Leggi quello che installi.
| Versione | Pubblicata | Commit | SHA-256 | Firma | Cambio di permessi |
|---|---|---|---|---|---|
| v2026.11.13-21cd568 | 13 nov 2026 | 21cd568 | ad066328 | Firmata | – |
Altro da AeonDave
Tutto da questo editoreAltro in Reverse engineering e fuzzing
Vai alla categoriaChiedi tutto quello che il tuo perimetro consente.
Crea un account, aggiungi crediti e fai la prima domanda in chat o nel terminale.
- 1/3
Crea l'account
Privato o azienda. Puoi passare ad azienda in seguito senza perdere saldo o chiavi.
PrivatoAzienda - 2/3
Aggiungi crediti
Con carta. Paghi a token, senza abbonamento.
$5.00ricarica minima - 3/3
Chiedi
In chat, oppure installa la CLI ed esegui notholy login nel terminale.
$ notholy login