Skip to content

hani-q/qstack / skills/qstack-next

qstack-next

Verified

Recommend the one installed skill to run next, from what the conversation has done so far and the state of the plan folder on disk. Use when invoked as /qstack-next, when asked "what now?" or which skill to use next, or as the last step of /qstack.

FileKindSize
SKILL.mdskill md3.7 KB
agents/openai.yamlconfig244 B
---
name: qstack-next
description: >
  Recommend the one installed skill to run next, from what the conversation
  has done so far and the state of the plan folder on disk. Use when invoked
  as /qstack-next, when asked "what now?" or which skill to use next, or as
  the last step of /qstack.
license: MIT
metadata:
  author: hani
  stack: qstack
---

# /qstack-next

Say what was just done, what that leaves undone, and the one skill that closes
the gap. Three or four lines. Name the skill; never run it.

## Evidence, heaviest first

1. **The plan folder.** Files on disk say where the work is more reliably than
   the conversation does. See the table below.
2. **The working tree.** `git branch --show-current`, `git status --short`,
   `git log --oneline -5`. Uncommitted changes mean work that is not yet
   verified or reviewed.
3. **The conversation.** What the user asked last, what was produced, what was
   verified. Facts, not tone: "it's broken" followed by three guards and no
   reproduction is evidence; a frustrated message is not.

Read what you need and no more. This skill is read-only.

## The plan lifecycle

Plan folders are `qstack/compound_engineering/plans/<slug>/`, or the legacy
`compound-engineering/plans/<slug>/`. The folder whose slug matches the branch
or the conversation is the one that matters. Its files arrive in a fixed
order, and the first one missing is the next step:

| The folder has | Run next |
| --- | --- |
| Nothing yet, and the user is describing a feature or change | `/qstack-plan-prior-art`, then draft `plan.md` |
| `plan.md` and no `plan.html` | `/qstack-plan-to-html` |
| `plan.html` with notes at `data-status="open"` | `/qstack-ask-plan-open-questions` |
| `plan.html` and no `board-events.js`, or a board without its complete-breakdown marker | `/qstack-plan-to-html` to finish the board |
| `board-events.js` with cards still open | the execution loop `execution.md` shows was used; `/qstack-loop-no-nonsense` when it shows none |
| every card closed and no `outcome.md`, on a board with no `review` epic | `/qstack-plan-adherence-review`, then `/qstack-plan-close` |
| every card closed and no `outcome.md`, on a board whose `review` epic gate card is `done` | `/qstack-plan-close`; the gate card already ran the adherence review |
| `outcome.md` | nothing from the lifecycle |

Work with no plan folder and no feature being described is outside the
lifecycle; do not push it in.

## Outside the lifecycle

Match the conversation against the `description:` of every installed skill.
The listing script in `/qstack` prints them; `ls ~/.claude/skills
~/.codex/skills ~/.agents/skills` names them. Each description says when its
skill applies; that text is the rule, so this file does not repeat it.

Prefer, in order: a QStack skill, a skill from a collection QStack's installer
offers, anything else installed. Never recommend a skill that is not installed.

## Output

Three or four short lines of plain prose, no heading, no list:

1. What was just done. One concrete fact: a file written, a card closed, a
   branch merged, a failure described.
2. What that leaves undone.
3. `Run /<skill>`, with its argument when the skill takes one. Name a second
   skill only when it follows directly, as `plan-close` follows
   `plan-adherence-review`.

When a file decided the recommendation, name the file. When nothing is undone
and no description matches: output nothing if `/qstack` called this skill,
otherwise one line, `Nothing pending.` Never pad to reach four lines.

## Hard rules

- Read-only. Edit nothing, run no skill, start no work.
- One recommendation, not a menu.
- Evidence, not mood. Every recommendation rests on a file, a diff, or a
  message you can point to.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 470kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall