ai-tools

Stakeholder requirements-intake skill — helps a stakeholder or user, via a series of structured prompts, turn their ideas or requests into a right-sized, refinable scope artifact. Adaptive: asks a few high-leverage questions when a request isn't specific enough, drafts-and-trims when it's too broad. Honest about what a non-technical stakeholder cannot know. Stand-alone; does not hand off to other agents.

Pull data out of raw files (PDF, XLSX, CSV/TSV, JSON, HTML, Markdown, plain text), clean each one, and converge everything into one consistent dataset ready for analysis. Use when you have a pile of messy files to extract and normalize. Scales from one file to many via an explicit parallelism ladder.

Answer a question or find patterns in a clean dataset — aggregations, groupings, comparisons, trends, outliers, distributions, joins. Use after data has been ingested and normalized, when the user wants insight rather than just extraction. Single-pass; produces findings, not a formatted report.

Deliver analysis findings as a polished output in one or more formats — Markdown, CSV, XLSX, JSON, plain text, or inline. Use as the final pipeline stage once analysis is done. Always asks for the format if unspecified, and always documents the artifacts and scripts it created.

Inbound request triage skill — the front door of the product work-shaping funnel. Classifies a raw, unsorted request into a routing bucket (product shaping, engineering, data, needs-info, or not-actionable) and produces a fast triage decision artifact. Single-pass; inline-only; stand-alone — recommends a next step but never invokes another agent or skill.

Engineering hand-off packaging skill — turns a refined scope brief into a self-contained, liftable engineering hand-off artifact that seeds the engineer agent's design phase. Produces a separate liftable artifact; stand-alone; points to the engineer agent but never invokes it.

Technical product-manager skill — matures a stakeholder scope artifact into a system-aware, ticket-ready brief in collaboration with engineering. Augments the same document in place with system considerations, authoritative scope, complexity, resolved open questions, and a ticket breakdown. Stand-alone; does not hand off to other agents.

Sub-agent implementer — executes individual tasks from a plan with TDD, self-review, and structured reporting. For orchestrators: dispatch this skill per independent task, then follow the review pipeline.

Write comprehensive implementation plans from approved designs — task decomposition with exact file paths, code, commands. Includes independence analysis for parallel dispatch and no-placeholder enforcement.

Complete development work — verify tests pass, detect environment state, present structured merge/PR options to user, execute choice with proper cleanup.

Mandatory pre-work step — explores project context, asks clarifying questions one at a time, proposes approaches, presents design for user approval. Terminal state: user-approved design doc ready for planning.

Dual-role review skill — handles both spec compliance review (did they build what was requested?) and code quality review (is it well-built?). Orchestrators dispatch this per completed task or for final cross-batch validation.