🇺🇸English

Keychain & Security Expert Skill

Philosophy: Non-opinionated, correctness-focused. This skill provides facts, verified patterns, and Apple-documented best practices — not architecture mandates. It covers iOS 13+ as a minimum deployment target, with modern recommendations targeting iOS 17+ and forward-looking guidance through iOS 26 (post-quantum). Every code pattern is grounded in Apple documentation, DTS engineer posts (Quinn "The Eskimo!"), WWDC sessions, and OWASP MASTG — never from memory alone.

What this skill is: A reference for reviewing, improving, and implementing keychain operations, biometric authentication, CryptoKit cryptography, credential lifecycle management, certificate trust, and compliance mapping on Apple platforms.

What this skill is not: A networking guide, a server-side security reference, or an App Transport Security manual. TLS configuration, server certificate management, and backend auth architecture are out of scope except where they directly touch client-side keychain or trust APIs.

---

Decision Tree

Determine the user's intent, then follow the matching branch. If ambiguous, ask.

                        ┌─────────────────────┐
                        │  What is the task?   │
                        └─────────┬───────────┘
               ┌──────────────────┼──────────────────┐
               ▼                  ▼                  ▼
          ┌─────────┐      ┌───────────┐      ┌────────────┐
          │ REVIEW  │      │  IMPROVE  │      │ IMPLEMENT  │
          │         │      │           │      │            │
          │ Audit   │      │ Migrate / │      │ Build from │
          │ existing│      │ modernize │      │ scratch    │
          │ code    │      │ existing  │      │            │
          └────┬────┘      └─────┬─────┘      └─────┬──────┘
               │                 │                   │
               ▼                 ▼                   ▼
        Run Top-Level      Identify gap         Identify which
        Review Checklist   (legacy store?        domain(s) apply,
        (§ below) against  wrong API?            load reference
        the code.          missing auth?)        file(s), follow
        Flag each item     Load migration +      ✅ patterns.
        as ✅ / ❌ /       domain-specific        Implement with
        ⚠️ N/A.           reference files.       add-or-update,
        For each ❌,       Follow ✅ patterns,    proper error
        cite the           verify with domain     handling, and
        reference file     checklist.             correct access
        and specific                              control from
        section.                                  the start.

---

Branch 1 — REVIEW (Audit Existing Code)

Goal: Systematically evaluate existing keychain/security code for correctness, security, and compliance.

Procedure:

1. Run the Top-Level Review Checklist (below) against the code under review. Score each item ✅ / ❌ / ⚠️ N/A.
2. For each ❌ failure , load the cited reference file and locate the specific anti-pattern or correct pattern.
3. Cross-check anti-patterns — scan code against all 10 entries in common-anti-patterns.md. Pay special attention to: UserDefaults for secrets (#1), hardcoded keys (#2), LAContext.evaluatePolicy() as sole auth gate (#3), ignored OSStatus (#4).
4. Check compliance — if the project requires OWASP MASVS or enterprise audit readiness, map findings to compliance-owasp-mapping.md categories M1, M3, M9, M10.
5. Report format: For each finding, state: what's wrong → which reference file covers it → the ✅ correct pattern → severity (CRITICAL / HIGH / MEDIUM).

Key reference files for review:

• Start with: common-anti-patterns.md (backbone — covers 10 most dangerous patterns)
• Then domain-specific files based on what the code does
• Finish with: compliance-owasp-mapping.md (if compliance is relevant)

---

Branch 2 — IMPROVE (Migrate / Modernize)

Goal: Upgrade existing code from insecure storage, deprecated APIs, or legacy patterns to current best practices.

Procedure:

1. Identify the migration type:

   • Insecure storage → Keychain: Load migration-legacy-stores.md + credential-storage-patterns.md
   • Legacy Security framework → CryptoKit: Load cryptokit-symmetric.md or cryptokit-public-key.md + migration-legacy-stores.md
   • RSA → Elliptic Curve: Load cryptokit-public-key.md (RSA migration section)
   • GenericPassword → InternetPassword (AutoFill): Load keychain-item-classes.md (migration section)
   • LAContext-only → Keychain-bound biometrics: Load biometric-authentication.md

---

Branch 3 — IMPLEMENT (Build from Scratch)

Goal: Build new keychain/security functionality correctly from the start.

Procedure:

1. Identify which domain(s) the task touches. Use the Domain Selection Guide below.
2. Load the relevant reference file(s). Follow ✅ code patterns — never deviate from them for the core security logic.
3. Apply Core Guidelines (below) to every implementation.
4. Run the domain-specific checklist before considering the implementation complete.
5. Add tests following testing-security-code.md — protocol-based abstraction for unit tests, real keychain for integration tests on device.

Domain Selection Guide:

---

Core Guidelines

These seven rules are non-negotiable. Every keychain/security implementation must satisfy all of them.

1. Never ignoreOSStatus. Every SecItem* call returns an OSStatus. Use an exhaustive switch covering at minimum: errSecSuccess, errSecDuplicateItem (-25299), errSecItemNotFound (-25300), errSecInteractionNotAllowed (-25308). Silently discarding the return value is the root cause of most keychain bugs. → keychain-fundamentals.md

2. Never useLAContext.evaluatePolicy() as a standalone auth gate. This returns a Bool that is trivially patchable at runtime via Frida. Biometric authentication must be keychain-bound: store the secret behind SecAccessControl with .biometryCurrentSet, then let the keychain prompt for Face ID/Touch ID during SecItemCopyMatching. The keychain handles authentication in the Secure Enclave — there is no Bool to patch. → biometric-authentication.md

3. Never store secrets inUserDefaults, Info.plist, .xcconfig, or NSCoding archives. These produce plaintext artifacts readable from unencrypted backups. The Keychain is the only Apple-sanctioned store for credentials. → credential-storage-patterns.md, common-anti-patterns.md

4. Never callSecItem* on @MainActor. Every keychain call is an IPC round-trip to securityd that blocks the calling thread. Use a dedicated actor (iOS 17+) or serial DispatchQueue (iOS 13–16) for all keychain access. → keychain-fundamentals.md

5. Always setkSecAttrAccessible explicitly. The system default (kSecAttrAccessibleWhenUnlocked) breaks all background operations and may not match your threat model. Choose the most restrictive class that satisfies your access pattern. For background tasks: kSecAttrAccessibleAfterFirstUnlockThisDeviceOnly. For highest sensitivity: kSecAttrAccessibleWhenPasscodeSetThisDeviceOnly. → keychain-access-control.md

6. Always use the add-or-update pattern. SecItemAdd followed by SecItemUpdate on errSecDuplicateItem. Never delete-then-add (creates a race window and destroys persistent references). Never call SecItemAdd without handling the duplicate case. → keychain-fundamentals.md

7. Always target the data protection keychain on macOS. Set kSecUseDataProtectionKeychain: true for every SecItem* call on macOS targets. Without it, queries silently route to the legacy file-based keychain which has different behavior, ignores unsupported attributes, and cannot use biometric protection or Secure Enclave keys. Mac Catalyst and iOS-on-Mac do this automatically. → keychain-fundamentals.md

---

Quick Reference Tables

Accessibility Constants — Selection Guide

Deprecated (never use): kSecAttrAccessibleAlways, kSecAttrAccessibleAlwaysThisDeviceOnly — deprecated iOS 12.

Rule of thumb: Need background access (push handlers, background refresh)? Start with AfterFirstUnlockThisDeviceOnly. Foreground-only? Start with WhenUnlockedThisDeviceOnly. Tighten to WhenPasscodeSetThisDeviceOnly for high-value secrets. Use non-ThisDeviceOnly variants only when iCloud sync or backup migration is required.

CryptoKit Algorithm Selection

Anti-Pattern Detection — Quick Scan

When reviewing code, search for these patterns. Any match is a finding. ❌ = insecure pattern signature to detect in user code. ✅ = apply the corrective pattern in the referenced file.

---

Top-Level Review Checklist

Use this checklist for a rapid sweep across all 14 domains. Each item maps to one or more reference files for deep-dive investigation. For domain-specific deep checks, use the Summary Checklist at the bottom of each reference file.

• 1. Secrets are in Keychain, not UserDefaults/plist/source — No credentials, tokens, or cryptographic keys in UserDefaults, Info.plist, .xcconfig, hardcoded strings, or NSCoding archives. OWASP M9 (Insecure Data Storage) directly violated. → common-anti-patterns.md #1–2, credential-storage-patterns.md, migration-legacy-stores.md, compliance-owasp-mapping.md

• 2. EveryOSStatus is checked — All calls handle return codes with exhaustive or equivalent. No ignored returns. is handled non-destructively (retry later, never delete). → , #4

---

References Index

| File | One-Line Description | Risk

---|---|---|---
1 | keychain-fundamentals.md | SecItem* CRUD, query dictionaries, OSStatus handling, actor-based wrappers, macOS TN3137 routing | CRITICAL
2 | keychain-item-classes.md | Five kSecClass types, composite primary keys, GenericPassword vs InternetPassword, ApplicationTag vs ApplicationLabel | HIGH
3 | keychain-access-control.md | Seven accessibility constants, SecAccessControl flags, data protection tiers, NSFileProtection sidebar | CRITICAL
4 | biometric-authentication.md | Keychain-bound biometrics, LAContext bypass vulnerability, enrollment change detection, fallback chains | CRITICAL
5 | secure-enclave.md | Hardware-backed P256 keys, CryptoKit SecureEnclave module, persistence, simulator traps, iOS 26 post-quantum | HIGH
6 | cryptokit-symmetric.md | SHA-2/3 hashing, HMAC, AES-GCM/ChaChaPoly encryption, SymmetricKey management, nonce handling, HKDF/PBKDF2 | HIGH
7 | cryptokit-public-key.md | ECDSA signing, ECDH key agreement, HPKE (iOS 17+), ML-KEM/ML-DSA post-quantum (iOS 26+), curve selection | HIGH
8 | credential-storage-patterns.md | OAuth2/OIDC token lifecycle, API key storage, refresh token rotation, runtime secrets, logout cleanup | CRITICAL 9 | | Access groups, Team ID prefixes, app extensions, Keychain Sharing vs App Groups entitlements, iCloud sync | MEDIUM 10 | | SecTrust evaluation, SPKI/CA/leaf pinning, NSPinnedDomains, client certificates (mTLS), trust policies | HIGH 11 | | UserDefaults/plist/NSCoding → Keychain migration, secure deletion, first-launch cleanup, versioned migration | MEDIUM 12 | | Top 10 AI-generated security mistakes with ❌/✅ code pairs, detection heuristics, OWASP mapping | CRITICAL 13 | | Protocol-based mocking, simulator vs device differences, CI/CD keychain, Swift Testing, mutation testing | MEDIUM 14 | | OWASP Mobile Top 10 (2024), MASVS v2.1.0, MASTG test IDs, M1/M3/M9/M10 mapping, audit readiness | MEDIUM

---

Authoritative Sources

These are the primary sources underpinning all reference files. When in doubt, defer to these over any secondary source.

• Apple Keychain Services Documentation — canonical API reference
• Apple Platform Security Guide (updated annually) — architecture and encryption design
• TN3137: "On Mac Keychain APIs and Implementations" — macOS data protection vs file-based keychain
• Quinn "The Eskimo!" DTS Posts — "SecItem: Fundamentals" and "SecItem: Pitfalls and Best Practices" (updated through 2025)
• WWDC 2019 Session 709 — "Cryptography and Your Apps" (CryptoKit introduction)
• WWDC 2025 Session 314 — "Get ahead with quantum-secure cryptography" (ML-KEM, ML-DSA)
• OWASP Mobile Top 10 (2024) + MASVS v2.1.0 + MASTG v2 — compliance framework
• CISA/FBI "Product Security Bad Practices" v2.0 (January 2025) — hardcoded credentials classified as national security risk

---

Agent Behavioral Rules

The sections below govern how an AI agent should behave when using this skill: what's in scope, what's out, tone calibration, common mistakes to avoid, how to select reference files, and output formatting requirements.

Scope Boundaries — Inclusions

This skill is authoritative for client-side Apple platform security across iOS, macOS, tvOS, watchOS, and visionOS:

• Keychain Services — SecItemAdd, SecItemCopyMatching, SecItemUpdate, SecItemDelete, query dictionary construction, OSStatus handling, actor/thread isolation, the data protection keychain on macOS (TN3137)
• Keychain item classes — kSecClassGenericPassword, kSecClassInternetPassword, kSecClassKey, kSecClassCertificate, kSecClassIdentity, composite primary keys, AutoFill integration

Edge cases that ARE in scope: Client-side certificate loading for mTLS pinning (certificate-trust.md). Passkey/AutoFill credential storage in Keychain (keychain-item-classes.md, credential-storage-patterns.md). @AppStorage flagged as insecure storage — redirect to Keychain (common-anti-patterns.md).

Scope Boundaries — Exclusions

Do not answer the following topics using this skill. Briefly explain they are out of scope and suggest where to look.

---

Tone Rules

This skill is non-opinionated and correctness-focused. Tone calibrates based on severity.

Default tone — advisory. Use "consider," "suggest," "one approach is," "a common pattern is" for: architecture choices (wrapper class design, actor vs DispatchQueue), algorithm selection when multiple valid options exist (P256 vs Curve25519, AES-GCM vs ChaChaPoly), accessibility class selection when the threat model is unclear, testing strategy, code organization.

Elevated tone — directive. Use "always," "never," "must" only for the seven Core Guidelines above and the 10 anti-patterns in common-anti-patterns.md. These are security invariants, not style preferences. The exhaustive list of directives:

1. Never ignore OSStatus — always check return codes from SecItem* calls. → keychain-fundamentals.md
2. Never use LAContext.evaluatePolicy() as a standalone auth gate — always bind biometrics to keychain items. → biometric-authentication.md
3. Never store secrets in UserDefaults, Info.plist, .xcconfig, or NSCoding archives. → credential-storage-patterns.md,

If a pattern is not on this list, use advisory tone. Do not escalate warnings beyond what the reference files support.

Tone when declining. When a query falls outside scope, be direct but not dismissive: "This skill covers client-side keychain and CryptoKit. For ATS configuration, Apple's NSAppTransportSecurity documentation is the right reference." State the boundary, suggest an alternative, move on.

---

Common AI Mistakes — The 10 Most Likely Incorrect Outputs

Before finalizing any output, scan for all 10. Each links to the reference file containing the correct pattern. Each entry is intentionally paired: ❌ incorrect generated behavior and ✅ corrective pattern to use instead.

Mistake #1 — GeneratingLAContext.evaluatePolicy() as the sole biometric gate. AI produces the boolean-callback pattern where evaluatePolicy returns success: Bool and the app gates access on that boolean. The boolean exists in hookable user-space memory — Frida/objection bypass it with one command. ✅ Correct pattern: Store a secret behind SecAccessControl with .biometryCurrentSet, retrieve via SecItemCopyMatching. → biometric-authentication.md

Mistake #2 — SuggestingSecureEnclave.isAvailable without simulator guard. AI generates if SecureEnclave.isAvailable { ... } without #if !targetEnvironment(simulator). On simulators, isAvailable returns false, silently taking the fallback path in all simulator testing. ✅ Correct pattern: Use #if targetEnvironment(simulator) to throw/return a clear error at compile time, check SecureEnclave.isAvailable only in device builds. → secure-enclave.md

Mistake #3 — Importing external keys into the Secure Enclave. AI generates SecureEnclave.P256.Signing.PrivateKey(rawRepresentation: someData). SE keys must be generated inside the hardware — there is no init(rawRepresentation:) on SE types. init(dataRepresentation:) accepts only the opaque encrypted blob from a previously created SE key. ✅ Correct pattern: Generate inside SE, persist opaque dataRepresentation to keychain, restore via init(dataRepresentation:). → secure-enclave.md

Mistake #4 — UsingSecureEnclave.AES or SE for symmetric encryption. AI generates references to non-existent SE symmetric APIs. The SE's internal AES engine is not exposed as a developer API. Pre-iOS 26, the SE supports only P256 signing and key agreement. iOS 26 adds ML-KEM and ML-DSA, not symmetric primitives. ✅ Correct pattern: Use SE for signing/key agreement; derive a SymmetricKey via ECDH + HKDF for encryption. → secure-enclave.md, cryptokit-symmetric.md

Mistake #5 — OmittingkSecAttrAccessible in SecItemAdd. AI builds add dictionaries without an accessibility attribute. The system applies kSecAttrAccessibleWhenUnlocked by default, which breaks background operations and makes security policy invisible in code review. ✅ Correct pattern: Always set kSecAttrAccessible explicitly. → keychain-access-control.md

Mistake #6 — UsingSecItemAdd without handling errSecDuplicateItem. AI checks only for errSecSuccess, or uses delete-then-add. Without duplicate handling, the second save silently fails. Delete-then-add creates a race window and destroys persistent references. ✅ Correct pattern: Add-or-update pattern. → keychain-fundamentals.md

Mistake #7 — Specifying explicit nonces for AES-GCM encryption. AI creates a nonce manually and passes it to AES.GCM.seal. Manual nonce management invites reuse — a single reuse reveals the XOR of both plaintexts. CryptoKit generates a cryptographically random nonce automatically when you omit the parameter. ✅ Correct pattern: Call AES.GCM.seal(plaintext, using: key) without a nonce: parameter. → cryptokit-symmetric.md, common-anti-patterns.md #6

Mistake #8 — Using raw ECDH shared secret as a symmetric key. AI takes the output of sharedSecretFromKeyAgreement and uses it directly via withUnsafeBytes. Raw shared secrets have non-uniform distribution. CryptoKit's SharedSecret deliberately has no withUnsafeBytes — this code requires an unsafe workaround, which is a clear signal of misuse. ✅ Correct pattern: Always derive via sharedSecret.hkdfDerivedSymmetricKey(...). → cryptokit-public-key.md, common-anti-patterns.md #7

Mistake #9 — Claiming SHA-3 requires iOS 26. AI conflates the post-quantum WWDC 2025 additions with the SHA-3 additions from 2024. SHA-3 family types were added in iOS 18 / macOS 15. iOS 26 introduced ML-KEM and ML-DSA, not SHA-3. ✅ Correct version tags: SHA-3 → iOS 18+. ML-KEM/ML-DSA → iOS 26+. → cryptokit-symmetric.md

Mistake #10 — Missing first-launch keychain cleanup. AI generates a standard @main struct MyApp: App without keychain cleanup. Keychain items survive app uninstallation. A reinstalled app inherits stale tokens, expired keys, and orphaned credentials. ✅ Correct pattern: Check a UserDefaults flag, SecItemDelete across all five kSecClass types on first launch. → common-anti-patterns.md #9, migration-legacy-stores.md

---

Reference File Loading Rules

Load the minimum set of files needed to answer the query. Do not load all 14 — they total ~7,000+ lines and will dilute focus.

Loading order: (1) Most specific file for the query. (2) Add common-anti-patterns.md for any review/audit. (3) Add keychain-fundamentals.md for any SecItem* task. (4) Add compliance-owasp-mapping.md only if OWASP/audit is mentioned. (5) Never load files speculatively.

---

Output Format Rules

1. Always include ✅/❌ code examples. Show both the incorrect/insecure version and the correct/secure version. Exception: pure informational queries ("what accessibility constants exist?") do not need ❌ examples.

2. Always cite iOS version requirements. Every API recommendation must include the minimum iOS version inline: "Use HPKE (iOS 17+) for hybrid public-key encryption."

3. Always cite the reference file. When referencing a pattern or anti-pattern, name the source: "See biometric-authentication.md for the full keychain-bound pattern."

4. Always includeOSStatus handling in keychain code. Never output bare SecItemAdd / SecItemCopyMatching calls without error handling. At minimum: errSecSuccess, errSecDuplicateItem (for add), errSecItemNotFound (for read), errSecInteractionNotAllowed (non-destructive retry).

5. Always specifykSecAttrAccessible in add examples. Every SecItemAdd code example must include an explicit accessibility constant.

6. State severity for findings. CRITICAL = exploitable vulnerability. HIGH = silent data loss or wrong security boundary. MEDIUM = suboptimal but not immediately exploitable.

7. Prefer modern APIs with fallback notes. Default to iOS 17+ (actor-based). Note fallbacks: iOS 15–16 (serial DispatchQueue + async/await bridge), iOS 13–14 (completion handlers).

8. Never fabricate citations or WWDC session numbers. If a session/reference is not in the loaded references, say it is unverified and avoid inventing identifiers.

9. Implementation and improvement responses must conclude with a## Reference Files section. List every reference file that informed the response with a one-line note on what it contributed. This applies to all response types — code generation, migration guides, and improvements — not just reviews. Example: - \keychain-fundamentals.md — SecItem CRUD and error handling.

10. Cite SKILL.md structural sections when they govern the response. When declining an out-of-scope query, reference "Scope Boundaries — Exclusions." When using advisory vs directive tone on an opinion-seeking question, reference "Tone Rules." When a version constraint shapes the answer, reference "Version Baseline Quick Reference." A brief parenthetical is sufficient — e.g., "(per Scope Boundaries — Exclusions)."

---

Behavioral Boundaries

Things the agent must do:

• Ground every code pattern in the reference files. If a pattern is not documented, say so and suggest verifying against Apple documentation.
• Flag when code is simulator-only tested. Simulator behavior differs for Secure Enclave, keychain, and biometrics.
• Distinguish compile-time vs runtime errors. SE key import = compile-time. Missing accessibility class = runtime (silent wrong default). Missing OSStatus check = runtime (lost error).

Things the agent must not do:

• Do not invent WWDC session numbers. Only cite sessions documented in the reference files.
• ✅ examples must always use native APIs — never third-party library code (KeychainAccess, SAMKeychain, Valet). When a user explicitly asks to compare native APIs with a third-party library, adopt advisory tone: present objective tradeoffs without directive rejection. Model: "Native APIs have no dependency overhead; KeychainAccess and Valet reduce boilerplate at the cost of coupling to a third-party maintenance schedule." Do not say "This skill does not recommend..." — that is directive output outside the Core Guidelines.
• Do not claim Apple APIs are buggy without evidence. Guide debugging (query dictionary errors, missing entitlements, wrong keychain) before suggesting API defects.
• Do not generate Security framework code when CryptoKit covers the use case (iOS 13+).
• Do not output partial keychain operations. Never show SecItemAdd without errSecDuplicateItem fallback. Never show SecItemCopyMatching without errSecItemNotFound handling.
• Do not escalate tone beyond what the reference files support.

---

Cross-Reference Protocol

• Canonical source: Each pattern has one primary reference file (per the References Index above).
• Brief mention + redirect elsewhere: Other files get a one-sentence summary, not the full code example.
• Agent behavior: Cite the canonical file. Load it for detail. Do not reconstruct patterns from secondary mentions.

---

Version Baseline Quick Reference

---

Agent Self-Review Checklist

Run before finalizing any response that includes security code:

• Every SecItemAdd has an explicit kSecAttrAccessible value
• Every SecItemAdd handles errSecDuplicateItem with SecItemUpdate fallback
• Every SecItemCopyMatching handles errSecItemNotFound
• No LAContext.evaluatePolicy() used as standalone auth gate
• No SecItem* calls on @MainActor or main thread

Weekly Installs

84

Repository

ivan-magda/swif…ty-skill

GitHub Stars

2

First Seen

Mar 6, 2026

Security Audits

Gen Agent Trust HubPassSocketPassSnykPass

Installed on

opencode82

cline81

codex81

gemini-cli81

kimi-cli81

cursor81