0x7067

Transform user feedback into structured Linear issues with AI-enhanced parsing for labels, priority, acceptance criteria, and estimates

Guidance on when and how to apply design patterns. Use when: (1) asking which pattern to use, (2) refactoring code, (3) discussing code smells, (4) need to decouple components, (5) building extensible systems.

Expert guidance on Swift Concurrency best practices, patterns, and implementation. Use when developers mention: (1) Swift Concurrency, async/await, actors, or tasks, (2) "use Swift Concurrency" or "modern concurrency patterns", (3) migrating to Swift 6, (4) data races or thread safety issues, (5) refactoring closures to async/await, (6) @MainActor, Sendable, or actor isolation, (7) concurrent code architecture or performance optimization.

Shared foundation for the sdlc-* skill library: the engineering standard every session must meet, and the .ai-sdlc/state.md / journal.md file spec for cross-session handoff. Load when any sdlc-* skill references it or when reading/writing those files.

End-of-work discipline for project repos: hostile validation of the diff before declaring work done, and handoff (settle the tree, update state.md, append the journal) before ending a session with changes. Triggered by the claim, not the diff — use before any done/ship verdict, even for small edits.

Start-of-work discipline for project repos: orientation before the first substantive task of a session, and planning whenever a task is multi-step, risky, ambiguous, or touches config that propagates beyond this machine. Execution is governed by sdlc-core.

Use when asked to improve the AIDP plugin itself by finding exactly one small maintainer-reviewable bug, docs ambiguity, confusing error, test gap, or contributor-experience papercut.

Use when running, reviewing, or auditing software work with AIDP v2 typed artifacts, bounded context, proof commands, and completion audits.

Closed feedback loop that hill-climbs the hashline harness. Each cycle re-measures a fixed baseline and a candidate back-to-back, compares them paired per-fixture, and keeps the candidate only if it improves the objective without regressing pass rate beyond noise. Use when the user asks to "optimize hashline", "run the improvement loop", or "tune the harness against the benchmark".

Use when working in Claude Code with Sonnet and Haiku as daily-driver coding models. Routes implementation to Sonnet, cheap scouting/log/diff work to Haiku, and keeps verification evidence tight.