<!-- Generated from models/_canonical by scripts/build-model-variants.js. Edit the canonical source, not this file. Behavioural profile for GLM: scripts/model-profiles.json -->

#Task boundary

1. Implement only what the task names; no extra abstractions or files.
2. English-only comments and identifiers.
3. Stop when the checklist passes.

---

#Purpose

A working pass over the OWASP Top 10 (2021 edition) categories. Each entry names the risk, shows what it looks like in real code, and points at the control.

This is a triage map, not a substitute for the deep packages. Where a category has one, follow the link — A03 alone spans Security/sql-injection, Security/xss and Security/command-injection.

---

#A01 — Broken Access Control

The most common category, and the one with the highest real-world impact.

An authenticated user reaches an object that is not theirs by changing an identifier. Route-level checks do not prevent it; the query must be scoped.

js
// Broken — the route is guarded, the row is not
const invoice = await db.invoice.findUnique({ where: { id: req.params.id } });

// Fixed — ownership is part of the lookup
const invoice = await db.invoice.findFirst({
  where: { id: req.params.id, organisationId: req.user.organisationId },
});

Controls: deny by default; scope every query; return 404 not 403; never accept role or tenant id from the client. → Security/authorization

#A02 — Cryptographic Failures

Sensitive data exposed through weak or absent cryptography.

Controls: TLS everywhere with HSTS; argon2id for passwords, never md5/sha1; AES-GCM or ChaCha20-Poly1305 for data, never ECB; keys in a KMS; never invent a scheme. → Security/encryption, Security/https

#A03 — Injection

Untrusted data parsed as code. SQL, OS commands, LDAP, XPath, template engines and XSS all sit here.

Controls: parameterised queries; argument arrays instead of shells; contextual output encoding; allow-list any dynamic identifier. → Security/sql-injection, Security/command-injection, Security/xss

#A04 — Insecure Design

A flaw in what was specified, not how it was built. No amount of correct implementation fixes a design that permits unlimited password resets or trusts a client-supplied price.

Controls: threat model before building; define abuse cases alongside use cases; enforce business limits server-side; assume every client is hostile.

#A05 — Security Misconfiguration

Defaults left in place, debug enabled in production, verbose errors, unnecessary features exposed.

ini
# Wrong — a stack trace tells an attacker your framework, version and paths
NODE_ENV=development

Controls: harden defaults; disable directory listing; strip stack traces from responses; remove sample apps and default accounts; review headers. → Security/headers

#A06 — Vulnerable and Outdated Components

The dependency with a known CVE that nobody upgraded.

bash
npm audit --omit=dev          # fail the build on high and critical

Controls: an SBOM; automated dependency updates; audit in CI; remove unused dependencies; track upstream advisories.

#A07 — Identification and Authentication Failures

Credential stuffing, weak recovery, session fixation, missing MFA.

Controls: breach-screen passwords; rate limit per account and per IP; rotate the session identifier on login; identical responses for unknown accounts; offer WebAuthn. → Security/authentication, Security/passwords

#A08 — Software and Data Integrity Failures

Trusting code or data whose provenance is unverified — unsigned updates, untrusted CI plugins, insecure deserialization.

js
// Never deserialize untrusted input into live objects
const data = JSON.parse(body);        // data, inert
// not: eval(body), or a deserializer that reconstructs arbitrary classes

Controls: lockfiles with integrity hashes; pin CI actions by commit SHA; sign artifacts; never deserialize untrusted input into executable objects.

#A09 — Security Logging and Monitoring Failures

The breach nobody noticed. Median dwell time is measured in months precisely because this control is missing.

Controls: log authentication outcomes, authorisation denials and privilege changes with subject, object and action; ship logs off-host; alert on anomalies; never log secrets, tokens or passwords. → Security/audit-log

#A10 — Server-Side Request Forgery

The server fetches a URL an attacker chose, reaching internal services the attacker cannot.

js
// The classic target: cloud instance metadata
// http://169.254.169.254/latest/meta-data/iam/security-credentials/

Controls: allow-list destination hosts; resolve DNS and check the resulting IP against private ranges — including 169.254.169.254, 127.0.0.0/8, 10/8, 172.16/12, 192.168/16; block redirects or re-validate each hop; require IMDSv2.

---

#Using this list

1. Treat it as coverage, not a ranking of your specific risk. Your threat model decides priority.
2. The Top 10 is a floor. Passing it is not a security programme.
3. Map each category to a test, not a document. A01 becomes a test that user B cannot read user A's invoice.

---

#Anti-patterns

---

#Checklist

• Every data access is scoped by owner or tenant in the query (A01)
• Passwords use argon2id; transport is TLS with HSTS (A02)
• All queries parameterised; all output contextually encoded (A03)
• Abuse cases defined and business limits enforced server-side (A04)
• Debug disabled, stack traces stripped, defaults hardened (A05)
• Dependency audit runs in CI and fails on high or critical (A06)
• Breach screening, rate limiting and session rotation in place (A07)
• Lockfiles with integrity hashes; CI actions pinned by SHA (A08)
• Auth outcomes and authorisation denials logged and shipped off-host (A09)
• Outbound fetches allow-listed and checked against private IP ranges (A10)