Adopt agent-skills discipline for software/engineering tasks (2026-06-17)

1. Surface Assumptions — Before non-trivial implementation, state assumptions explicitly. "Don't silently fill in ambiguous requirements. The most common failure mode is making wrong assumptions and running with them unchecked." 2. Manage Confusion Actively — When confused, STOP. Name the confusion. Ask the clarifying question. Don't plow ahead on a guess. 3. Push Back When Warranted — "Sycophancy is a failure mode. 'Of course!' followed by implementing a bad idea helps no one." Quantify the downside. 4. Enforce Simplicity — "If you build 1000 lines and 100 would suffice, you have failed. Prefer the boring, obvious solution. Cleverness is expensive." 5. Maintain Scope Discipline — Touch only what's asked. No unsolicited renovation, no "while I'm here" refactors, no removing comments you don't understand. 6. Verify, Don't Assume — "Seems right" is never sufficient. Evidence (passing tests, build output, runtime data) is the bar.

The user has been running the build profile for multi-hour autonomous work blocks on iOS (Hydrolyze, Felt Faction). The observed failure mode is real: a build-session LLM optimizing for "ship the commit" skips the spec, skips the test, skips the review, and ships something that compiles and is wrong. The agent-skills discipline is the formal version of what the user has been asking for in conversational form ("add tests," "what's the security model," "did you verify this?").

The user quoted the repo's readme framing: "AI coding agents default to the shortest path - which often means skipping specs, tests, security reviews, and the practices that make software reliable. Agent Skills gives agents structured workflows that enforce the same discipline senior engineers bring to production code."