Design Compensation Strategy (design_compensation_strategy)

Use this skill when a Blueprint or JSON architecture contains mutating steps and you need to add robust, LIFO-ordered compensation logic and corresponding test cases.

When to Use

• •Adding addCompensation blocks to a BaseBlueprint or architecture
• •Identifying non-reversible actions and defining HITL or manual remediation
• •Matching "Undo" capabilities from the MCP Registry for Connectors/Commanders
• •Writing or extending blueprint tests for failure-and-rollback scenarios

Instructions

1. •Mutation analysis: Scan for this.execute() calls that are not read-only (Connectors, Commanders). Categorize as CREATE, UPDATE, DELETE, PROVISION.
2. •Semantic matching: Query MCP Registry for Undo capabilities (e.g., create_issue → delete_issue). If no match, ask the user (HITL) for compensation tool ID or strategy.
3. •Non-reversible guardrails: For "point of no return" actions (e.g., payments, physical world), require a manual gate or compensation log step for human remediation. Do not ignore.
4. •Code injection: Insert addCompensation immediately after the successful step; capture IDs from step output for the compensation call. Ensure compensations are idempotent and use the same runAs context.
5. •Tests: Add negative scenarios: force step failure, exhaust retries, assert all prior compensations were called (test-driven-development).
6. •Output: Provide Saga design summary, updated Blueprint code with compensations, and updated blueprint.test.ts with rollback assertions.

For the full system prompt, see references/design-compensation-strategy.md.