tkersey/dotfiles
Overview
This skill clarifies ambiguous or conflicting requests by researching first, then asking only the judgment calls needed to converge on a concrete definition of the problem. It operates in a Discover+Define loop: gather discoverable facts, produce a one-line problem statement plus measurable success criteria, and stop before any implementation. Use it when prompts request pressure-testing, hard questions, or system-design decisions prior to building.
How this skill works
I inspect available project facts and maintain a Snapshot (stage, problem statement, success criteria, facts, decisions, open questions). I research discoverable information before asking anything, then ask tight batches of 1–3 judgment questions (prefer 2) until there are no blocking open questions. When in high-pressure mode I force concreteness (metrics, dates, owners) and re-ask vague answers using stable question ids.
When to use it
- Prompt includes "$grill-me", "grill me", or similar phrasing
- You need scope, success criteria, or acceptance signals before implementation
- You want assumptions pressure-tested or trade-offs surfaced
- A request asks for system-design or optimization decisions before coding
- You need a concise handoff-ready problem definition
Best practices
- Research first: do not ask for discoverable facts; inspect artifacts and update Snapshot
- Ask only judgment calls: prefer 2 independent questions per batch; use 1 only for ordered dependencies
- Keep questions concrete: include metrics, dates, scope boundaries, and an owner when applicable
- Use stable snake_case ids and short headers for follow-ups; re-ask with the same id for clarifications
- Stop at a fully defined Snapshot: one-line problem statement, measurable success criteria, and no blocking open questions
Example use cases
- Clarify ambiguous feature requests before scoping implementation work
- Pressure-test product goals and prioritize trade-offs with concrete metrics
- Define success criteria and non-goals for an optimization project
- Convert a loosely specified system-design prompt into a single-line problem and acceptance signal
- Prepare a design handoff package that other skills can implement
FAQ
I prefer 2 questions per batch, up to 3 when independent; use 1 only when sequence is required.
What happens if a user answer is vague?
I re-ask the same question id demanding concreteness (metric/date/owner) until the Snapshot is concrete.
19 skills
This skill researches first, then clarifies scope with judgment-based questions, delivering a concrete problem statement and measurable success criteria before
This skill generates Deckset markdown presentations from conversation context, preserving claims, structure, and presenter notes for cohesive narrative slides.
This skill lets you automate and orchestrate the codex app-server via a v2 proxy, streaming JSONL with deterministic timeouts.
This skill creates or refactors Codex skills with minimal diffs, updating frontmatter, SKILL.md, and agents/openai.yaml while keeping lean.
This skill helps you manage Codex automations by editing the local SQLite store and filesystem configs, enabling fast creation, updates, and runs.
This skill enforces inductive invariants at boundary points to prevent impossible states and improve reliability across systems.
This skill translates git-like intents to non-interactive xit cli commands, avoiding the TUI and using --cli where available.
This skill captures evidence-backed execution learnings and persists them as JSONL in .learnings.jsonl for reuse in future turns.
This skill enables automated control of the codex app-server via a v2 proxy, streaming JSONL, auto approvals, and multi-instance orchestration.
This skill tightens wording and compresses text while preserving meaning, boosting clarity and scan-ability across prompts, docs, and emails.
This skill finalizes validated work by confirming signals, summarizing proof, optionally commenting beads, and opening a PR without merging.
This skill automates PR orchestration using gh commands to keep PRs green and ready for human handoff.
This skill orchestrates swarm subagents to optimize solution quality through structured critique and consensus-driven task completion.
This skill turns a plan into a dependency-aware DAG in SLICES.md, validating and selecting the next ready slice to execute.
This skill aggressively improves latency, throughput, and memory usage by profiling, measuring, and applying safe algorithmic optimizations with guards.
This skill helps you reduce incidental code complexity by providing an analysis-first plan for readability, with actionable simplification steps and TRACE
This skill finalizes a GitHub PR end-to-end by updating branches, monitoring checks, squash-merging, and cleaning local/remote state.
This skill refines an existing Codex skill using minimal diffs and quick validation to improve reliability and triggers.
This skill helps you persist and review durable task plans in your repo with JSONL, ensuring continuity across sessions.
