Audit rule catalog

All 19 audit rules: what each detects, severity, confidence, false-positive notes and how to fix.

Version 0.1.0 · Source: docs/handbook/12-rule-catalog.md

Severity and confidence are the defaults emitted by the rule. IDs are stable; the #NNN suffix in a finding is positional per run.

Rule	Title	Severity	Confidence	Category
AS-SEC-001	Possible hard-coded credential	critical	medium	security
AS-SEC-002	Private key material committed	critical	high	security
AS-SEC-003	Environment file present in repository tree	high	medium	security
AS-SEC-004	Dynamic code or shell execution	high	medium	security
AS-SEC-005	SQL built by string concatenation or interpolation	high	medium	security
AS-SEC-006	TLS certificate verification disabled	high	high	security
AS-IAC-001	Over-permissive infrastructure configuration	medium	medium	infrastructure
AS-ARC-001	Circular dependency between modules	medium	high	architecture
AS-ARC-002	Lower layer depends on a higher layer	medium	medium	architecture
AS-ARC-003	Very large source files	low	medium	maintainability
AS-OPS-001	No CI pipeline configuration found	medium	medium	operability
AS-OPS-002	No test files found	medium	medium	maintainability
AS-OPS-003	Container runs as root	medium	high	security
AS-SUP-001	Dependency manifest without lockfile	medium	high	supply-chain
AS-SUP-002	Unpinned dependency ranges (`latest`, `*`)	medium	high	supply-chain
AS-SUP-003	Container base image not pinned	low	medium	supply-chain
AS-AI-001	LLM call co-located with dynamic code execution	high	low	ai-security
AS-DOC-001	No README at repository root	low	high	developer-experience
AS-DOC-002	No architecture decision records (only if more than 200 source files)	info	medium	governance

Details#

AS-SEC-001 Possible hard-coded credential#

Detects: AWS access key IDs, GitHub tokens, sk- style keys, Slack tokens, Google API keys, and quoted password|secret|api_key|token|access_key assignments of 8+ characters.
Skips: test and fixture paths, .md/.txt, lines mentioning example/placeholder/changeme/your_/xxxx or reading process.env, os.environ, getenv.
Output: kind and a redacted excerpt. The value is never printed.
Fix: revoke and rotate first; then move to a secret manager or environment; purge history only after rotation.
Limits: misses unusual formats; may flag test-like values outside test paths.

AS-SEC-002 Private key material committed#

Detects: a BEGIN ... PRIVATE KEY header followed by base64 key material (next line, or on the same line). Regex literals and docs that merely mention the header do not trigger.
Fix: treat as compromised; rotate; remove.

AS-SEC-003 Environment file in tree#

Detects: .env, .env.* files not named *.example, *.sample, *.template. Contents are never read into the report.
Limits: reads the working tree, not git tracking; a git-ignored local .env is still listed.

AS-SEC-004 Dynamic code or shell execution#

Detects: eval(, new Function(, child_process exec with interpolated or concatenated strings, os.system(, Python subprocess with shell=True.
Skips: tests, docs, non-code files.
Fix: argument arrays, parsers, allowlists.
Limits: cannot prove tainted input.

AS-SEC-005 SQL string building#

Detects: query|execute|exec|raw|prepare( called with a template literal containing ${ or string concatenation; and uppercase SQL statements (SELECT .. FROM, INSERT INTO, UPDATE .. SET, DELETE FROM) combined with interpolation or concatenation on the same line.
Fix: bind parameters.
Limits: line-based; multi-line queries can be missed; constants may be flagged.

AS-SEC-006 TLS verification disabled#

Detects: rejectUnauthorized: false, verify=False, InsecureSkipVerify: true, NODE_TLS_REJECT_UNAUTHORIZED=0.
Limits: may be a dev-only profile.

AS-IAC-001 Over-permissive infrastructure#

Detects: Terraform cidr_blocks = ["0.0.0.0/0"]; Kubernetes privileged: true, allowPrivilegeEscalation: true, hostNetwork: true.
Limits: public load balancers on 80/443 are legitimate.

AS-ARC-001 Circular dependency#

Detects: strongly connected components of size > 1 in the module graph.
Limits: module = top-level directory heuristic.

AS-ARC-002 Layer inversion#

Detects: a file in a service/domain layer importing from presentation, or a data/persistence layer importing from service or presentation, using folder-name conventions (controllers/routes/handlers/api/views/pages/ui/web; services/usecases/application/domain; repositories/dao/db/database/models/persistence/infrastructure).
Limits: relies on naming conventions.

AS-ARC-003 Very large files#

Detects: non-test source files over 1000 lines.
Limits: generated code and tables.

AS-OPS-001 No CI#

Detects: none of GitHub Actions, GitLab CI, Azure Pipelines, Jenkinsfile, CircleCI. Evidence kind is unknown.

AS-OPS-002 No tests#

Detects: no test/, tests/, __tests__, spec, e2e, *.test.*, *.spec.*, *_test.go|py. Evidence kind is unknown.

AS-OPS-003 Container runs as root#

Detects: a Dockerfile with no USER, or whose last USER is root/0.

AS-SUP-001 Missing lockfile#

Detects: package.json with dependencies and no package-lock.json, pnpm-lock.yaml, yarn.lock, bun.lock(b) beside it or at the root.

AS-SUP-002 Unpinned ranges#

Detects: dependency version latest or * in package.json.

AS-SUP-003 Unpinned base image#

Detects: FROM image:latest or FROM image without tag or digest.

AS-AI-001 LLM plus execution sink#

Detects: a file that both calls an LLM API (OpenAI/Anthropic-style identifiers) and contains an AS-SEC-004 sink. Confidence is low because data flow is not proven.
Fix: never execute model output; allowlisted tools with validated arguments, sandboxing, human approval.

AS-DOC-001 / AS-DOC-002 Documentation gaps#

Detects: missing README; missing ADR directory in large codebases.

What the audit does not do#

It does not execute code, query live systems, scan networks, inspect cloud accounts, read git history, or assess runtime behavior. Use the skills and your own judgment to cover those areas.