Claude Skills by DYAI2025

Use when turning a feature idea into an implementation-ready PRD with traceable REQ IDs, acceptance criteria, NFRs, risks, data model, and task slices for /agileteam Phase 0.

Use when a requirement, design gap, blocker, or open question must be closed with the user instead of being guessed by the agent.

Use during retrospectives to discover recurring process, prompt, skill, or workflow improvement candidates before authoring persistent changes.

Use when designing verification strategy so independent gates cover requirements, implementation, security, validation, and judgment without claiming perfect certainty.

Use while carrying out an approved implementation plan, keeping task state, evidence, and deviations explicit.

Checks every external claim for evidence before the final answer and classifies it as supported, inferable, unverified, or do-not-claim. Companion to ultrathink-craftsmanship. Use when stating releases, APIs, prices, benchmarks, studies, product behavior, legal/medical/financial facts, claims about foreign codebases/files/org knowledge, or any numbers, names, dates, versions, quotes — and before any autonomous agent flow adopts a claim that could propagate.

Use optionally before ai-native-prd-architect when a vague product idea needs a user-facing feature spec and scope framing.

Use after the same bug signature recurs two or more times so fixes stop treating symptoms and identify the causal chain before further edits.

Use when creating or updating a Claude Code skill directory with a focused SKILL.md, safe trigger conditions, workflow, guardrails, and validation notes.

Use when a verification gate fails and the next fix needs reproducible diagnosis instead of speculative patching.

Use when implementing a code task so the test is written or updated before production code and proves the requested behavior.

Use when reviewing tests for fragility, over-mocking, false positives, missing production wiring, or assertions that do not prove the requirement.

Use for expensive, one-pass reasoning gates that stress-test a specification, plan, or near-final implementation for bias, hidden coupling, weak evidence, and craftsmanship risks.

Use before non-trivial work on a default branch to isolate changes in a feature branch or dedicated git worktree.

Use when creating implementation plans that can be executed by humans or agents with traceable requirements and validation commands.

Use when authoring or modifying a Claude Code skill so the result is focused, discoverable, safe, and not duplicative.