Skill Details
Runtime
(claude-4-7-opus)
Example user message
[userMessage]: Add a scheduled job that emails each organization a weekly summary of their activity.
Example agent response
Plan First
Entering plan mode. No edits until the plan below is approved.
1) Isolation Model
- Tenant identifier source: organizations registry iteration.
- Propagation channel: scoped context set per iteration; finally-block cleanup.
- Enforcement boundary: repository wrapper for activity reads; templating layer for email composition is tenant-scoped.
- Failure mode: skip the organization with an audit log when context cannot be set.
- Audit signal: per-iteration structured log with org id and email send result.
2) Changed Access Paths
- LIST activity for the iteration's org : list : authority `system:background-reader`
- WRITE outbound email row : write : authority `system:notifier`
3) Invariants Applied
- No global activity query; iteration is explicit.
- Email content composed inside the tenant context; templates do not read other-tenant data.
- Logs include the org id but not user-level identifiers.
- Cleanup runs on success and on failure.
4) Tests Added
- Negative read: scoping wrapper rejects an unscoped activity query.
- Negative write: notifier rejects writes without tenant context.
- Negative list: per-iteration query returns only the iteration's tenant rows.
- Context-missing: iteration without context is skipped and logged.
- Async propagation: helpers inside the iteration inherit the tenant context.
5) Gate Results
- T1: pass.
- T2: pass.
- T3: pass.
- T4: pass.
- T5: pass.
- T6: pass.
6) Residual Risks
- Risk: long iterations may delay other jobs; surface a metric for iteration duration.
7) Next Actions
- Add a follow-up to expose a skipped-iteration metric to the dashboard.
- Add a follow-up to verify email rendering does not include cross-tenant attachments.
Get this skill + 9 more with SelectFrom $14/mo · Cancel anytimeFrom $14/mo · Cancel anytime
View SelectBy purchasing this skill, you agree to our terms of service
CLAUDE-4-7-OPUS
AI-edited tenant-aware backends silently leak data across organizations. This is the Cursor-native self-contained skill: drop the .mdc into .cursor/rules/ and the Agent enforces isolation invariants, scoping-boundary rules, authority mapping, tenant-negative test obligations, and 6 audit gates. Includes skill identifier, when-to-use, setup, trigger conditions, forbidden behaviors, two worked examples, and an audit-friendly response contract. A matching Claude Code edition is sold separately.
...more
Added over 1 month ago
