Menu
Scan the Stella codebase for security issues. We handle sensitive legal documents — security is non-negotiable.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
| name | security-audit |
| description | Scan the Stella codebase for security issues. We handle sensitive legal documents — security is non-negotiable. |
Scan the Stella codebase for security issues. We handle sensitive legal documents — security is non-negotiable.
Hardcoded secrets — scan for API keys, tokens, passwords, connection
strings in source files (not .env.example or .env):
Search for patterns like apiKey, secret, password, token,
credential assigned to string literals in apps/ and packages/.
Ignore markdown, example files, and node_modules.
Auth bypass — verify that every handler in
apps/api/src/handlers/ checks authentication. Read each handler's
route file and confirm auth middleware is applied. Flag any endpoint
that is publicly accessible without explicit justification.
Ownership via branded types — every workspace-scoped handler
must use createHandler from api-handlers.ts, which provides a
WorkspaceHandlerContext with workspaceId: SafeId<"workspace">.
Flag any handler that:
workspaceId from the request body or query params
instead of the validated contextcreateRootHandler when it accesses workspace-scoped datastring where SafeId<"workspace">,
SafeId<"organization">, or SafeId<"user"> is expectedpermissions declaration in its configS3 presigned URLs — read apps/api/src/handlers/files/presign.ts
and any other presign usage. Check that:
../)File upload validation — check that uploaded files are validated for type and size before being stored in S3. Verify that file metadata (content type, name) from the client is not blindly trusted.
CORS configuration — read apps/api/src/index.ts and check the
CORS setup. Flag origin: "*" or overly permissive origins. Should
be restricted to your known domains.
Dependency vulnerabilities:
If bunfig.toml configures a security scanner, run the configured
Bun dependency scan:
bun pm scan
If no scanner is configured, note that local Bun dependency scanning is unavailable for this repo and check the Dependabot alerts instead:
gh api repos/{owner}/{repo}/dependabot/alerts --jq '.[].security_advisory.summary'
AI prompt injection — check that user-provided content passed to
the AI SDK (Google Generative AI) is not interpolated directly into
system prompts. Read the AI integration code in apps/api/src/ and
verify that user content is clearly separated from system
instructions.
Workspace isolation — handlers must use scopedDb (which
pre-filters by workspace) for all workspace-scoped queries. Flag
any handler that:
db directly for workspace-scoped tablesWHERE workspaceId = ... instead of using
scopedDbEthical walls (information barriers) — in law firms, lawyers working on opposing sides of a matter (or for competing clients) must not have access to each other's documents. Verify that:
Auth and session management — the project uses better-auth with
organisation support, roles (Owner, Admin, Member), and email OTP.
Check apps/api/src/lib/auth.ts and apps/api/src/db/auth-schema.ts.
Verify that:
Data retention and deletion — when a document or matter is deleted, verify that:
Legal privilege leakage — attorney-client privilege and work product doctrine mean document content must never leak through side channels:
Cross-matter data isolation in AI context — when AI features (review, research, chat) build context:
searchMatter, readEntity, etc.) must
enforce matter-level permissions per call, not per
sessionDocument versioning integrity — legal documents require an unbroken chain of custody:
Audit trail completeness — SOC 2 and ISO 27001 require evidence of who did what and when:
Sensitive data in logs and errors — legal data requires stricter controls than typical SaaS:
Atomicity on privileged operations — operations that affect access to privileged data must be atomic:
For each finding, report:
Group findings by severity. If no issues found in a category, confirm it passed.
If critical or high severity issues are found, fix them immediately and
create a commit: fix: address security audit findings.
Drive the stella command-line client (@stll/cli), a legal-workspace CLI whose command surface is generated from the stella MCP tool registry. Covers install, OAuth login, the full command tree grouped by domain, JSON output for scripting, the --input escape hatch for deep payloads, cursor pagination, destructive-op confirmation, and exit codes.
Apply when writing or reviewing React effects in apps/web. Direct useEffect is banned; use the sanctioned wrappers or a better primitive.
Apply when writing or modifying database schema, queries, or migrations.
Apply when a performance-guard check (network baseline, bundle baseline, DB query count, loader-prefetch lint, RC bailouts) fails or when touching a hot route/endpoint.
Internationalization conventions for Stella. Apply when adding or modifying user-facing strings.
Apply when building or modifying user-facing UI components.