AI Agent Workflow Collection

Use when about to call the askquestion (or equivalent) tool, or when considering whether to ask the user a question. Teaches a four-gate decision procedure: trigger, fit, construct, validate. Addresses under-trigger (LLM doesn't ask unless prompted), mis-appropriation (LLM uses the tool for open-ended questions), and prose fallback as a last resort. Load before any tool-mediated user clarification.

Produces a code implementation plan by grilling the user on technical choices (language, framework, dependencies, project structure, sub-projects, project type) before tickets are created. Use only when a spec or PRD is already present — referenced as a file path, attached as a document, or substantively laid out in the conversation — for a code/programming project. Defer to domain-grilling for vague ideas, domain modeling, or terminology alignment. Do not use for general planning, non-code/non-programming projects, or creating a spec/PRD itself.

A relentless interviewing skill focused on Domain-Driven Design (DDD) alignment. Use for vague ideas with no spec or PRD, domain modeling, terminology alignment, or conceptual design before any spec exists. Walks through the design tree iteratively, resolves terms against a glossary (CONTEXT.md), and documents critical architectural decisions (ADRs). Defer to code-implementation-grilling once a spec or PRD is present and technical decisions (language, framework, dependencies, project structure) are the goal.

Coordinate implementation of a normalized ticket set — builds the dependency DAG, groups same-file tickets, dispatches each group to a sub-agent with a fresh context, runs a judge LLM against acceptance criteria, and commits one commit per ticket. Use when a batch of tickets should be implemented in parallel/sequential order with per-ticket commits and an end-of-run report. Do not use for a single ticket (implement it directly), tickets not yet generated (use spec-to-tickets first), or non-ticket work.

Assess whether a project is ready for its FIRST public release (initial release or 1.0) by exploring the codebase, running the build and tests, and triaging findings into Blockers, Should-haves, and Nice-to-haves. Use when a user asks if a project is ready to ship, cut a 1.0, or do an initial release. Not for post-1.0 patch/minor releases or routine maintenance.

Configures a host repository to use the alastairlundy AI-agent skill family by writing four agent-consumable docs (issue-tracker, triage-labels, domain/GLOSSARY, decision-ledger-audit) and adding the Agent skills block to AGENTS.md or CLAUDE.md. Use when setting up a new repo for AI agents, switching the repo's issue tracker, or re-running after a config change.

Create implementation tickets with dependency graphs, Independent/Collaborative classification, and context pointers from a spec, PRD, or conversation context. Output to issue tracker or local markdown. Use when - spec/PRD exists and is complete, need tickets sized for focused work sessions (3-4 hours each), want to enable parallel work with explicit dependency tracking, need Independent/Collaborative classification per ticket. Don't use when - spec is incomplete or vague (use domain-grilling or to-prd first), need different granularity like epics or tasks, want to implement directly without decomposition, user explicitly wants to send tickets to issue tracker AND doesn't need dependency graphs or classification (use to-issues instead).

Generates a repo-agnostic, user-facing markdown changelog by analyzing git history, transforming commit messages, and categorizing changes into logical sub-projects.