Work Prep

Work Prep

Full preparation workflow: ticket context gathering followed by implementation planning.

This skill orchestrates two existing skills in sequence. It does not duplicate their logic.

When to Use

• User shares a ticket key (e.g. SKU-123, ST-456, CORE-789)
• User provides a Jira/Atlassian URL
• User says "start work on", "pick up ticket", "plan this feature", "work prep", "let's work on [ticket]"
• User pastes a ticket key with no other context

Workflow

Step 0: Lock the goal (anti-stall)

Hard requirement. Before invoking any sub-skill, you (the executing model) MUST call /goal yourself. work-prep is a multi-skill chain (intake → atdd-plan → confirm) that emits a large context brief and a large plan mid-flow, both are points where the model tends to read "turn complete" and stop. A /goal is a session-scoped Stop-hook: after every turn a fast evaluator checks your condition against the conversation and makes you continue if it is unmet. Set it here, once, spanning the whole chain. (Grounded in docs/research/2026-06-11-claude-code-goal-anti-stall.md.)

Phrase the condition as something your own output demonstrates, the evaluator cannot run tools or read files, only judge what you have surfaced:

A codebase-grounded context brief has been produced; the most blocking clarifying questions have been asked via the AskUserQuestion tool and answered (or none were blocking); /flagrare:atdd-plan has produced an implementation plan with acceptance tests; and the closing next-step AskUserQuestion (Start implementation / Adjust the plan / Stop here) has been presented and answered. Stop after 20 turns if not met.

One goal per session. Because work-prep sets this spanning goal, /flagrare:intake must NOT set its own, the [work-prep] prefix in Step 1 signals it to skip (a second goal would silently replace this one). Likewise do not re-set the goal between steps.

If /goal is unavailable (untrusted workspace, or disableAllHooks / allowManagedHooksOnly set): proceed without it and rely on the same-turn handoffs and the sub-skills' own no-yield notes.

Also create a Todo list (TodoWrite) with one item per stage of the chain.

Step 1: Invoke /flagrare:intake

Call /flagrare:intake with the ticket reference prefixed by [work-prep] (e.g., args: [work-prep] SKU-123). This prefix tells intake two things: (1) skip setting its own /goal, work-prep's Step 0 goal already spans the chain, and (2) skip its Step 6 next-step prompt and hand off directly. This skill will:

1. Parse the ticket ID/URL and identify the platform
2. Read the full ticket via MCP (Jira, Linear, etc.)
3. Follow all referenced links in parallel (Notion, Figma, Confluence, GitHub, etc.)
4. Synthesise a context brief (without questions yet)
5. Ground the brief in the codebase via /flagrare:codebase-explore, finding the files, utilities, and prior attempts the plan will touch
6. Ask codebase-informed clarifying questions (specific: "extend src/x.ts or fork it?", not abstract: "where should this live?")
7. Resolve open questions with the user

Wait for /flagrare:intake to complete before proceeding. The context brief must be finalized, codebase findings populated, and open questions resolved.

Step 2: Invoke /flagrare:atdd-plan

Once /flagrare:intake has produced a complete, codebase-grounded context brief, invoke /flagrare:atdd-plan. Pass the brief as opening context. This skill will:

1. Run its own /flagrare:codebase-explore pass (atdd-plan stays self-sufficient, intake's findings in the brief are additional input, not a substitute)
2. Produce an ATDD-first implementation plan with acceptance tests, named design patterns, SOLID audit, and gap review
3. Present the plan for user review

Step 3: Confirm readiness: use the AskUserQuestion tool

After the plan is presented, issue an AskUserQuestion tool call, same interaction shape as intake's next-step prompt and plan-mode's accept tool. Do NOT phrase this as prose; that lets the turn end ambiguously.

Options:

• Start implementation (Recommended): proceed to write code against the plan.
• Adjust the plan: collect specific changes from the user, re-run the relevant atdd-plan steps, then re-present.
• Stop here: return control with the plan saved/printed for later.

Anti-patterns

• Don't skip the Step 0 /goal. It's the forcing function that carries the model across the brief and the plan without stopping. The [work-prep] prefix on intake assumes this goal exists.
• Don't let a sub-skill set its own goal. Only one goal is active per session; a second silently replaces work-prep's spanning goal and the chain loses its anti-stall guarantee mid-flow.
• Don't skip /flagrare:intake and jump to planning. Context gaps turn into rework.
• Don't invoke /flagrare:atdd-plan before clarifying questions are resolved.
• Don't start implementation before the plan is reviewed and approved.
• Don't re-fetch context that /flagrare:intake already gathered. Pass the brief forward.
• Don't ask clarifying questions before codebase grounding. Questions asked without knowing what exists are abstract and frequently miss the real ambiguity. intake handles this by calling /flagrare:codebase-explore before Step 5.
• Don't be surprised that /flagrare:codebase-explore runs twice in this flow, once in intake (scoped to inform questions) and once in atdd-plan (scoped to anchor the plan). atdd-plan stays standalone-callable; that requires it to do its own pass.
• Don't end work-prep with a prose "what next?" question. After the plan is presented, close with a tool-driven AskUserQuestion prompt, same UX contract as plan-mode's accept tool. A button, not a typing prompt.

Critical: The handoff mechanism

The [work-prep] prefix in intake's args is the signal. When intake sees it, intake skips its next-step prompt and immediately invokes /flagrare:atdd-plan itself. This means:

• Work-prep invokes intake (with prefix) → intake runs its full workflow → intake invokes atdd-plan → atdd-plan produces plan → work-prep resumes at Step 3.
• Work-prep does NOT need to invoke atdd-plan itself. Intake handles the handoff when it sees the [work-prep] prefix.

If you are executing work-prep and intake finishes WITHOUT invoking atdd-plan (e.g., it returned with just the brief), you MUST invoke /flagrare:atdd-plan yourself before proceeding to Step 3.

Flow position

/flagrare:work-prep [ticket ID or URL]
     |
     v
[Step 0: /goal locked, spans the whole chain, anti-stall]
     |
     v
/flagrare:intake [work-prep] <- 0. skip own /goal (work-prep owns it)
                                1. read ticket + follow references in parallel
                                2. synthesise brief (no questions yet)
                                3. /flagrare:codebase-explore  <- ground brief in code
                                4. ask codebase-informed clarifying questions
                                5. resolve, finalize brief
                                6. detect [work-prep] prefix → invoke atdd-plan directly
     |
     v
/flagrare:atdd-plan       <- runs its own /flagrare:codebase-explore pass
                             (stays self-sufficient for standalone use),
                             then produces ATDD-first plan
     |
     v
[user reviews plan]
     |
     v
[implementation begins]