Agent Skills: What Next

Use when deciding what to work on next from the current thread or the surrounding project. Triggers on "what next", "what should we do next", or "where were we".

UncategorizedID: v1-io/v1tamins/v1-what-next

Install this agent skill to your local

pnpm dlx add-skill https://github.com/v1-io/v1tamins/tree/HEAD/plugins/v1tamins/skills/v1-what-next

Skill Files

Browse the full folder contents for v1-what-next.

Download Skill

Loading file tree…

plugins/v1tamins/skills/v1-what-next/SKILL.md

Skill Metadata

Name
v1-what-next
Description
Use when deciding what to work on next from the current thread or the surrounding project. Triggers on "what next", "what should we do next", or "where were we".

What Next

Name the one next thing to do.

Start from this thread. Look outside it only when the thread has no clear next move or one current fact in a named source decides that move. If neither the thread nor the project shows one, say so. Don't guess. Recommend the action; do not start implementation or mutate a tracker from this skill.

If they want a v1 skill, not a next step, use v1-menu. If what's needed is a plan, point at a planning skill instead of writing the plan here.

Workflow

1. Reorient

From the thread, name the original goal in one sentence, what's been verified as done, and the open loops: unanswered questions, steps that were proposed but never taken, errors seen and not resolved. Leave decisions the user already made alone.

Run git status only when the next step depends on whether work is committed and the thread doesn't already say. Do not expand into the backlog when the thread already establishes the next move. If the choice depends on one current fact in a named task, plan, PR, or handoff, check that source read-only through the authorized route. Stop once the deciding fact is resolved. If access is unavailable, name the source gap separately from a decision only the user can make; do not ask the user to invent current source state.

2. Look outward when the thread runs dry

If the goal is done, or nothing left in the thread ranks above "everything else", find where this project keeps follow-up work before you answer. Don't assume a particular tool. Look in this order:

  • Whatever the thread named as the owner of the work: a ticket, plan, or task it started from.
  • The project's instructions or contributor docs, which usually say where work is tracked.
  • What the repo or workspace actually has: an issue list, a plans or tasks directory, a roadmap, a TODO file, open review requests.

Read only enough to find the item closest to what just happened: same area, same goal, or a follow-up the finished work created. Stop at the first source that gives you candidates. Mark anything found this way as outside the thread so the user knows where it came from. If nothing turns up, say that and offer the smallest wrap-up that won't get lost.

3. Rank

Pick the next move in this order:

  1. Finishes or unblocks the original goal.
  2. Verifies something currently claimed but unproven.
  3. Prevents loss: uncommitted work, an unrecorded decision, an answer that would be expensive to reconstruct later.
  4. Everything else, including follow-ups and related work found outside the thread.

Do the smallest thing that moves the top item. If that item is waiting on a user decision, the next move is that decision, asked simply, with a default. If the last response left several questions, name the one that actually blocks and default the rest.

4. Answer

**Next:** [one concrete action, one sentence]

**Why:** [one or two sentences tying it to the goal and state; mark anything
unverified]

**Also on the table:** [only when a runner-up is a genuinely close call]

**Needs you:** [only when a decision blocks the next step: the question and
a default]

Don't offer a menu of equals. When the work is done, say so, and say whether to stop, commit, or hand it off.

Example

The last response ended with four questions about naming, file layout, test scope, and rollout. Answer: "Next: confirm the file layout. I default to option A for naming and skip rollout for now; the other questions do not block implementation."