Built-in Detections

Pacakaged Rules

AKA ships seven rule packs out of the box — 101 detection rules in total. Every pack is bundled into the plugin and installed active by default, under the log-only Monitor policy until you assign a stronger action (Warn / Redact / Block) to a detection from the dashboard. All rules include positive and negative fixtures that run as part of the test suite.

Pack updates are manual. The plugin scans with your installed snapshot of each pack (enabled packs only). Upgrading the plugin/CLI records the newer packs as available but changes nothing until you apply the update — from the dashboard's Detections page (Update), with aka detections update, or after reviewing /aka:detections in Claude Code. New packs added in a release are auto-installed (monitor-only); packs you already have are never modified automatically, and your enabled/policy choices survive every update.

core-pii

Location: rules/core-pii/

Detects personally identifiable information commonly found in prompts. 14 rules.

Rule ID	Severity	Matcher	Description
email	medium	regex	Standard email addresses (user@example.com)
ssn	high	regex	US Social Security Numbers (123-45-6789)
phone-us	high	regex	US phone numbers incl. +1 and area codes
phone-intl	high	regex	International E.164 numbers
ip-address	low	regex	IPv4 addresses
dob	high	regex	Dates of birth (1900–2019)
drivers-license-us	high	regex	US driver's license numbers (requires license/DL context)
passport-us	high	regex	US passport numbers (A12345678)
home-address	medium	keyword	Contextual address phrases
name	medium	keyword	Contextual name introduction phrases
vin	medium	regex	Vehicle Identification Numbers (17 chars)
mac-address	low	regex	MAC addresses (AA:BB:CC:DD:EE:FF)
national-id	high	regex	Multi-country: Aadhaar, UK NI, Canadian SIN
zip	low	regex	US ZIP / ZIP+4 postal codes

core-financial

Location: rules/core-financial/

Detects financial account information. 8 rules.

Rule ID	Severity	Matcher	Post-Validators	Description
credit-card	critical	regex	luhn	Visa, MC, Amex, Discover card numbers
iban	high	regex	—	International Bank Account Numbers
routing-number	high	regex	—	US ABA routing numbers (9 digits, requires routing/ABA/RTN context)
swift	high	regex	—	SWIFT/BIC codes (8 or 11 chars, requires SWIFT/BIC context)
cvv	high	regex	—	Card security codes (contextual)
cusip	medium	regex	—	US securities identifiers (requires CUSIP context)
paypal	medium	regex	—	PayPal transaction IDs (requires PayPal context)
salary	low	keyword	—	Salary/compensation contextual phrases

secrets

Location: rules/secrets/

Detects credentials and tokens that should never be shared with AI models. 21 rules.

Rule ID	Severity	Matcher	Post-Validators	Description
aws-access-key	critical	regex	entropy	AWS Access Key IDs (AKIA/ASIA/AROA/etc.)
aws-secret-key	critical	regex	entropy	AWS Secret Access Keys (40-char, requires aws-secret-key context)
gcp-service-account	critical	regex	—	GCP service account emails
azure-connection-string	critical	regex	—	Azure Storage/Service Bus/Cosmos connection strings (two keys)
openai-api-key	critical	regex	entropy	OpenAI API keys (sk-proj-..., sk-...)
anthropic-api-key	critical	regex	entropy	Anthropic API keys (sk-ant-...)
stripe-live-key	critical	regex	entropy	Stripe live keys (sk_live_, pk_live_)
slack-token	critical	regex	—	Slack tokens (xoxb-, xoxp-, xapp-)
discord-token	critical	regex	entropy	Discord bot tokens (3-part format)
sendgrid-key	critical	regex	entropy	SendGrid API keys (SG.)
twilio-key	critical	regex	—	Twilio Account SIDs (AC...)
gitlab-token	critical	regex	entropy	GitLab PATs (glpat-)
digitalocean-token	critical	regex	—	DigitalOcean PATs (dop_v1_)
datadog-key	critical	regex	entropy	Datadog API/APP keys
npm-token	critical	regex	entropy	npm access tokens (npm_)
heroku-api-key	critical	regex	—	Heroku API keys (UUID, requires heroku context)
cloudflare-api-key	critical	regex	entropy	Cloudflare API tokens (40-char, requires cloudflare context)
pulumi-access-token	critical	regex	entropy	Pulumi access tokens (pul-)
terraform-cloud-token	critical	regex	entropy	Terraform Cloud user tokens
vault-token	critical	regex	entropy	HashiCorp Vault tokens (hvs., s.)
github-pat	critical	regex	—	GitHub PATs/OAuth/App tokens

secrets-infra

Location: rules/secrets-infra/

Detects infrastructure-level secrets and credentials. 13 rules.

Rule ID	Severity	Matcher	Post-Validators	Description
ssh-private-key	critical	regex	—	RSA/DSA/EC/OpenSSH private key blocks
db-connection-string	critical	regex	—	DB URLs with credentials (postgresql://user:pass@...)
jwt-token	high	regex	entropy	JSON Web Tokens (3-part base64url)
basic-auth-url	critical	regex	—	HTTP URLs with embedded credentials
basic-auth-header	critical	regex	—	HTTP Basic auth Authorization headers
pgp-private-key	critical	regex	—	PGP private key blocks
bearer-token	high	regex	entropy	Generic Authorization Bearer tokens
env-key-value	high	keyword	—	Sensitive env vars with values
password-field	high	keyword	—	JSON/YAML password/secret/token fields
docker-config-auth	critical	regex	—	Docker config.json base64 auth
kubeconfig-token	critical	regex	—	Kubernetes kubeconfig embedded tokens
api-key-header	high	regex	entropy	API keys in x-api-key/api-key/apikey headers
generic-high-entropy-secret	high	regex	entropy	High-entropy values on credential-shaped keys (fallback)

core-phi

Location: rules/core-phi/

Detects protected health information (HIPAA-relevant). 8 rules.

Rule ID	Severity	Matcher	Description
mrn	high	regex	Medical Record Numbers (contextual)
hipaa-identifier	high	regex	Health plan beneficiary / claim numbers
member-id	high	regex	Health plan member IDs
group-id	high	regex	Health plan group IDs
icd10-code	medium	regex	ICD-10 diagnosis codes (requires ICD/diagnosis/dx context)
ndc-code	medium	regex	National Drug Codes
genetic-data	high	keyword	DNA/genetic data references
biometric-ref	high	keyword	Biometric data references

core-code-context

Location: rules/core-code-context/

Detects internal code and infrastructure context that should not leave the organization. 8 rules.

Rule ID	Severity	Matcher	Description
internal-ip	low	regex	RFC 1918 private IPs (10.x, 172.16-31.x, 192.168.x)
localhost-ref	low	regex	localhost, 127.0.0.1, ::1 references
internal-domain	low	regex	.internal, .corp, .local domains
file-path	low	regex	Absolute Unix (known roots) / Windows file paths
stack-trace	low	regex	Stack traces revealing internal structure
internal-url	low	keyword	Dev/staging environment URLs
feature-flag	low	keyword	Feature flag / toggle names
db-table-name	low	keyword	Internal SQL table names

code-flaws

Location: rules/code-flaws/

Detects insecure code patterns across Python, JS/TS, Java, Ruby, PHP, and .NET. Rules fire on both code written by Claude (PreToolUse on Write/Edit) and existing code read from disk (PostToolUse on Read). 29 rules covering OWASP Top 10 and related weaknesses.

Rule ID	Severity	Matcher	Description
sql-inject-concat	high	regex	SQL queries built with + string concatenation
sql-inject-concat-dot	high	regex	SQL queries built with PHP . string concatenation (PHP files only)
sql-inject-format	high	regex	SQL queries via Python % format or f-strings
sql-inject-interp	high	regex	SQL queries via Ruby #{} or PHP $var interpolation
cmd-inject-shell	critical	regex	subprocess.*shell=True near user-controlled input
cmd-inject-exec	critical	regex	Java Runtime.getRuntime().exec() near user input
cmd-inject-node-exec	critical	regex	Node.js exec/execSync with non-literal argument near user input
xss-inner-html	high	keyword	.innerHTML = / .innerHTML= assignments
xss-dangerously-set	high	keyword	React dangerouslySetInnerHTML
xss-unescaped-render	high	regex	raw(), .html_safe, \| safe, mark_safe() in templates
deser-pickle	critical	regex	Python pickle.loads() / pickle.load()
deser-yaml-unsafe	high	regex	yaml.load() without SafeLoader
deser-java-ois	critical	regex	Java ObjectInputStream + readObject()
hardcoded-password	high	regex	password/passwd/pwd assigned a high-entropy string literal
hardcoded-secret-key	high	regex	secret_key/api_key/auth_token assigned a high-entropy string literal
dev-debug-enabled	medium	regex	DEBUG = True / debug: true in config
dev-placeholder-secret	high	keyword	Placeholder strings (changeme, your-secret, etc.) near key/token fields
dev-wildcard-cors	medium	regex	CORS wildcard: Access-Control-Allow-Origin: *, origin: '*', CORS_ALLOW_ALL_ORIGINS = True
auth-ssl-verify-false	high	regex	requests.*verify=False disabling TLS validation
auth-jwt-no-verify	critical	regex	JWT decoded with verify=False / algorithms=[] near jwt/decode/token
path-traversal-open	high	regex	open(request.* / open(params.* with user-controlled path
path-traversal-join	high	regex	os.path.join(base, user_*) with user-controlled component
crypto-weak-hash-md5	medium	regex	MD5 used as a hash (hashlib.md5, MessageDigest("MD5"), Digest::MD5)
crypto-weak-hash-sha1	medium	regex	SHA-1 used near password/credential/token context
crypto-insecure-random	medium	regex	Math.random() / random.random() near security-sensitive labels
prototype-pollution-merge	high	regex	Object.assign(target, req.body) / _.merge(target, request.*)
eval-dynamic-exec	critical	regex	eval/exec with non-literal argument near user input
ssrf-user-url	high	regex	fetch/axios/requests.get called directly with req.*/params.*
regex-redos-backtrack	medium	regex	Catastrophic backtracking patterns like (a+)+

config-posture

Structural rules over the plugin's configuration-inventory scan (skills & hooks), evaluated once per session on SessionStart. Unlike the text packs above, these compare hook entries against each other, so they live as pure functions in packages/detections/src/posture/ (fixture-tested like every pack) rather than in the regex rule format. Findings are observational (warn) and power the hook status/warning lines on the config read surface.

Rule	Severity	Fires when
hook-conflict	medium	Two hooks on the same event with overlapping matchers both mutate the tool target — run order is undefined
hook-unknown	medium	A settings-scope hook's command is attributable to no installed plugin and no recognizable tool (v1 heuristic)
hook-external-egress	high	A hook command contains network-egress primitives (curl, wget, nc, ssh, …) or a URL

Running the test suite

pnpm --filter @akasecurity/detections test

# Watch mode during rule development
pnpm --filter @akasecurity/detections test -- --watch

Fixture tests auto-discover across all packs - no need to register new rules manually.

Adding your own rules

See Writing Rules for the full guide. Create a new directory under rules/, add a manifest.json, your rule JSON files, and fixtures/ — the test runner and pack loader auto-discover everything.