planlisted
Install: claude install-skill x-mesh/xm
# x-plan
## Planning Principles
- Keep the plan proportional to the work. Prefer the smallest approach that satisfies the user's actual goal; do not inflate a local change into a framework, migration, compatibility layer, or lifecycle.
- Treat the request as a hypothesis. Verify that the requested method achieves the underlying goal, check whether the repository already provides it, and surface a simpler adequate alternative when one exists.
- Do not invent fallbacks. A fallback belongs in the plan only when a concrete failure condition is evidenced, explicit failure would be worse, activation is observable, behavior differences are documented, and both paths have a test. Otherwise require a clear failure.
- Separate verified repository facts, inferences, and unresolved user-owned decisions. Never make a plan executable by guessing paths, APIs, validation commands, or risk assumptions.
- Identify what the change can break, then select only the smallest existing validation that directly observes that risk. Do not list test, lint, build, and review as a fixed checklist. Explain any intentionally omitted check. Propose a new gate only for a named failure that existing checks miss and only with a measurable unique-catch criterion.
- Enumerate failure modes for every risk-domain requirement (parsing, matching/regex, caching, concurrency/locking, queues, auth, crypto, input handling, streaming, protocol), and say how the code should behave when the limit is reached — not just how