Guides collaborative requirements discovery before implementation. Creates task directory, seeds PRD, asks high-value questions one at a time, researches technical choices, and converges on MVP scope. Use when requirements are unclear, there are multiple valid approaches, or the user describes a new feature or complex task.
71
88%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Passed
No findings from the security scan
A request to build, implement, fix, refactor, or "go ahead" is not approval to leave planning. Task-creation consent is also not implementation approval.
For every non-trivial task, the user must respond at least once after the initial request before implementation begins. If no clarification is needed, that response must approve the final planning summary described below.
While any user-owned product, scope, UX, compatibility, risk, or acceptance decision remains unresolved, end the turn with exactly one highest-value question. Do not edit product code, dispatch implementation, or run task.py start.
If a question can be answered by exploring the codebase, explore the codebase instead.
This is mandatory. Before asking the user a question, first check whether the answer is already available in code, tests, configs, docs, existing specs, or task history.
Do not ask the user to confirm facts that the repository can answer. Ask only for product intent, preference, scope, risk tolerance, acceptance behavior, or decisions that remain ambiguous after inspection.
Repository evidence establishes current behavior and technical constraints. The user's intended behavior, feature scope boundaries, and UX preferences are never answerable by repository evidence alone, even when an existing pattern exists; existing patterns are options and recommendation evidence, not decisions.
Use this skill during Phase 1 planning to turn the user's request into clear requirements and planning artifacts.
Use this skill only after task-creation consent has been given and the user is ready to enter Trellis planning.
If no task exists yet, create one:
TASK_DIR=$(python3 ./.trellis/scripts/task.py create "<short task title>" --slug <slug>)Use a concise title from the user's request. Use a slug without a date prefix. task.py create adds the MM-DD- directory prefix automatically.
task.py create creates the default prd.md. Update that file with the current understanding before asking follow-up questions.
prd.md.prd.md, recompute the decision inventory, and repeat from step 2.design.md and implement.md for complex tasks.task.py start or edit product code in the same turn.task.py start and implementation. If the artifacts change materially after approval, repeat the final review.Do not invent a project-specific product/spec hierarchy. If the repository already has product, domain, or spec docs, use them. If it does not, proceed with the evidence that exists.
Ask only one question per message.
Each question must include:
Do not ask process questions such as whether to search, inspect files, or continue brainstorming. Do the evidence work directly. Ask the user only when the remaining issue is a product decision, preference, scope boundary, or risk tolerance choice.
Recommendations are not default selections. Never choose a recommended product decision on the user's behalf merely because the user asked for implementation.
Do not manufacture clarification questions when the request and repository evidence already resolve every decision. In that case, proceed directly to the final planning summary, which still requires a subsequent explicit approval.
The final review is a required phase-transition gate, not a prohibited process question. Task-creation consent, the initial implementation request, and approval given before the latest final summary do not satisfy this gate.
When requirements are vague, solutions feel over-engineered, or you're about to add complexity "because everyone does" — decompose to fundamental truths before reasoning upward.
Strip away implementation details to one sentence.
Bad: "We need to add Redis caching to the user profile endpoint" Good: "User profile data takes too long to load"
What is absolutely true (not opinion or convention)?
| Category | Examples |
|---|---|
| Physical constraints | Network latency ≥ 0, disk I/O has limits |
| Business rules | "Users must see their own data" |
| Technical invariants | "Data must be consistent" |
| User needs | "The user wants X within Y seconds" |
For each component of the current plan:
Before final review, verify all of the following:
Lightweight tasks may omit design.md and implement.md; they may not skip evidence inspection, requirement convergence, final review, or fresh implementation approval.
The final planning summary must show Goal, In Scope, Out of Scope, Acceptance Criteria, Key Decisions, relevant Risks or Deferred Items, and artifact status.
prd.md records requirements and acceptance:
design.md records technical design for complex tasks:
implement.md records execution planning for complex tasks:
task.py startLightweight tasks may have only prd.md. Complex tasks must have prd.md, design.md, and implement.md before task.py start.
implement.md is not a replacement for implement.jsonl. On sub-agent-dispatch workflows, implement.jsonl and check.jsonl must each contain at least one real spec/research entry before task.py start; the seed _example row does not count. Inline workflows skip this JSONL gate because Phase 2 loads context through trellis-before-dev.
Before declaring planning ready or running task.py start, rewrite prd.md once against the final structure described in the artifact rules above. This is not optional cleanup; it is the final planning gate.
The pass must be lossless:
What I already know, Assumptions, and resolved Open Questions into Goal, Background, Requirements, Technical Notes, or Acceptance Criteria.After the pass, read prd.md top to bottom and verify that no fact is repeated across sections unless the repetition adds new information.
Before declaring planning ready:
prd.md contains testable acceptance criteria.prd.md has passed the PRD convergence pass: no unresolved temporary brainstorm sections, no duplicate facts across sections, and no lost anchors, decisions, or acceptance mappings.design.md and implement.md.implement.jsonl and check.jsonl; seed-only manifests are not ready.Do not start implementation merely because the user originally asked for implementation.
12e279a
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.