Developer docs/Guides

Addon security review checklist

Record reviewer, addon/package hash, source revision, SDK/ABI, host version, provider API version and evidence for each relevant item. Mark exclusions and residual risks explicitly. Passing the local harness is regression evidence, not approval, certification or a trust decision.

#Integration admission and maintenance

Do not treat addon count as the success metric. Before adding or promoting an integration, record:

• Named customer requests or anonymized demand evidence, required workflows and regional/payment requirements.
• Provider sandbox availability, API stability, rate limits, commercial/access requirements and deprecation policy.
• A maintainer, supported provider API versions, fixture coverage, a real-sandbox validation date and an incident/credential-rotation owner.
• The minimum privileges needed and whether an existing maintained integration already satisfies the workflow.
• Upgrade, reconciliation, webhook replay and operator recovery evidence appropriate to the integration's risk.

Prioritize blockers to actual customer migrations and frequently used revenue/provisioning workflows. Label packages as experimental, fixture-tested or sandbox-validated according to evidence; do not call an integration certified merely because it compiles or passes local fixtures. Stripe and DigitalOcean are representative fixture examples, not a claim that they are the highest-demand providers or production-certified. Review maintenance cost and unresolved incidents before admitting more integrations.

#Trust and release provenance

• [ ] Review source and dependency changes, reproducible build inputs, lockfiles, licenses, vulnerabilities, and the exact WASM/package selected for deployment.
• [ ] Validate package structure, sizes, paths and hashes before extraction. Verify signatures against operator-owned trusted keys before installation, migration, upgrade or activation; never trust a key bundled with an upload.
• [ ] Keep signing private keys out of source/packages/logs and backups accessible to addon code. Restrict filesystem permissions/Windows ACLs and signing access.
• ] Follow [signed addon releases. Record the signer/key ID, distribute new public keys over an authenticated channel, define the overlap period, revoke retired/compromised keys, and test unknown/revoked/invalid keys.
• [ ] Define emergency response for already-installed code after revocation. Signature validity alone does not prevent rollback or revoke a running addon. Document unsigned-package policy and downgrade controls in the application.

#Least privilege and isolation

• [ ] Request only required permissions and capabilities; review changes against the previously approved version. Approval is an application/operator decision, not something the SDK constants or harness grant.
• [ ] Prefer exact outbound hosts. Review $server, $setting:..., suffix allowlists, schemes, ports, redirects, private networks and cloud metadata access. Keep http_private and wildcard access off unless explicitly justified. Exercise real SSRF/DNS-rebinding controls outside the offline harness.
• [ ] Minimize declared secret names. Read secrets through host imports, not ordinary settings or JSON wire input. Never log auth headers, webhook signing material, passwords, tokens, private keys or complete customer/provider payloads.
• [ ] Test missing/revoked/rotated credentials and redaction. Do not rely on host redaction as permission to log secrets; transformed/split/short values can leak.
• [ ] Avoid fs, db/sql, exec, bridge, jobs and broad UI powers where possible. If necessary, review each path/table/program/argument/quota and runtime/operator allowlist; prefer structured commands over shell interpolation. Check symlinks, traversal, subprocess output, untrusted configuration and data exfiltration.
• [ ] Bound fuel, memory, wall time, HTTP payload sizes, call counts, queues and retry rates. Verify failure/cancellation does not leave a partial privileged action silently accepted. Check hooks/mixins/routes for auth bypass and recursion.

#Gateway and webhook behavior

• [ ] Check exact decimal/currency handling and minor-unit rounding; use stable payment/reference metadata and idempotency keys, including refund keys.
• [ ] Verify webhook signatures over original bytes with constant-time checks, enforce timestamp tolerance where the provider supports it, and test malformed, forged, expired and rotated-secret signatures. Do not authenticate by source URL or an unverified event field alone.
• [ ] Deduplicate verified event IDs durably in the application; handle reordered events, duplicate deliveries and concurrent callbacks. A parser returning the same event twice is not application deduplication.
• [ ] Reconcile amount, currency, provider account and local payment ownership before applying financial effects. Verify refunds, partial refunds and unknown statuses do not accidentally settle a payment. Limit sensitive error detail.
• [ ] Classify provider/network failures accurately; retry only safe operations with bounded backoff and reconciliation after ambiguous outcomes.

#Provisioning lifecycle and idempotency

• [ ] Validate remote IDs and template fields before constructing paths, queries, commands or JSON. Encode correctly and bind remote resources to the right customer/service/server; reject cross-tenant IDs.
• [ ] Persist the created remote ID/module data and reuse them for retries, suspend/resume and terminate. Test create replay with the saved ID, plus the harder crash-after-provider-success/before-local-commit window separately.
• [ ] Use provider idempotency support or explicit reconciliation for ambiguous creates. Do not advertise exactly-once creation from a local saved-ID shortcut.
• [ ] Document asynchronous create/action completion, polling, failed actions, suspend/resume ordering, already-absent termination, timeout and retry behavior. Do not interpret an accepted action as a completed one.

#Migrations, upgrades and operator recovery

These are review requirements for the application upgrade owner, not behavior implemented by the SDK or local harness.

• [ ] Inventory schema/data changes and enforce addon ownership/namespacing. Review raw SQL, grants, indexes, locks, destructive steps and access to core data.
• [ ] Verify trust before executing migration code. Reapproval must cover expanded capabilities; check every install/upgrade entry point, not only a CLI path.
• [ ] Define ordered migration history/checksums and retry/idempotency behavior. Test duplicate execution, partial failure, concurrent upgrades and interrupted activation. Where transactional DDL is unavailable, specify recovery steps.
• [ ] Test an upgrade from a real prior package/data snapshot, not only a clean install. Preserve customer/service IDs, money, provider references and secrets.
• [ ] Back up and test restore. Declare whether rollback is supported after each migration and whether old code can read new data; do not assume downgrading signed WASM reverses a schema migration.
• [ ] Separate package verification, extraction/staging, migration and activation. Document their ordering/atomicity and prevent requests/jobs using incompatible code/schema during cutover. Include credential/key rotation during upgrade.
• ] Record compatibility evidence and known limitations using [the compatibility contract. Keep deployment approval, residual-risk acceptance and operational monitoring separate from test output.