AIssistLab

Privacy Security Release Checklist

Use this checklist before any V1 checkpoint commit or release.

Local Access

• All endpoints that expose local filesystem, provider, env, profile, or index state are localhost-only.
• Non-local Host headers return 403.
• No endpoint performs broad filesystem scans.
• Folder browsing is constrained to explicit user-selected roots.
• Import flows preview before writing.

Secret Safety

• API keys are never serialized in API responses.
• API key fields are masked in UI.
• Settings API responses return redacted placeholders for secret-like env values and preserve existing private values when unchanged placeholders are saved.
• Chat stream errors are sanitized before they are sent to the browser.
• Child-process env values are never returned to the browser.
• Provider status output is sanitized before display.
• Diagnostics exports report present/missing/malformed only.

Claude Profile Privacy

• Public APIs do not include raw Claude profile paths.
• Public APIs do not include profile folder basenames when those basenames may identify an account or organization.
• Public APIs use generic profile labels.
• Manual paths are sanitized before display.
• Smoke-test state is keyed by provider/profile/config fingerprint and does not expose raw config paths.

Filesystem Safety

• Skill names reject path traversal.
• Skill writes stay inside the configured skills directory.
• Delete requires exact name confirmation.
• Future import flows write only after user confirmation.
• Future restore/trash flows keep backups inside an app-controlled local directory.

Claude Ecosystem Safety

• Claude ecosystem inventory is read-only in V1.
• The app does not auto-install skills, hooks, agents, MCP servers, or plugins.
• The app does not auto-run Claude Code slash commands.
• Hook/event observability is not enabled by default.
• Suggested fixes are text only unless a future spec explicitly adds a safe apply action.

Docs Hygiene

• Docs do not include personal account names.
• Docs do not include raw local home paths.
• Docs do not include sample secret prefixes.
• Docs do not include raw auth callback paths or token file paths.
• Third-party source text is paraphrased and cited.
• Leaked/internal Claude material is not used.

Verification Commands

Run after docs-only changes:

git diff --check
git status --short --branch

Run after code changes:

npm run verify:release
npm run release:evidence -- --gate-result passed

The release verifier includes full tests, lint, build, browser/API smoke, safe button smoke, cleanup dry-run checks for processes and local artifacts, asset/docs/dead-code/unused-export audits, tracked diff whitespace checks, untracked release-text hygiene checks, and the privacy scan.

For focused debugging, run individual checks:

npx --yes tsx src/lib/claude/discovery.test.ts
npx --yes tsx src/lib/rag/index-state.test.ts
npx --yes tsx src/lib/skills/guided-builder.test.ts
npx --yes tsx src/lib/skills/validation.test.ts
npx --yes tsx src/lib/skills/quality.test.ts
npx --yes tsx src/lib/settings/runtime-config.test.ts
npx --yes tsx src/lib/settings/doctor.test.ts
npx --yes tsx src/app/api/index/route.test.ts
npx --yes tsx src/app/api/skills/guided/routes.test.ts
npx --yes tsx src/app/api/skills/validation-route.test.ts
npx --yes tsx src/app/api/chat/status/route.test.ts
npx --yes tsx src/app/api/export/zip/route.test.ts
npm run lint
cmd.exe /c npm run build
npm audit