garrytan / garrytan/gstack

/autoplan needs a --tier (smoke/standard/full) selector for non-production runs

Open
#1,955 1 comment 0 reactions 0 assignees View on GitHub
Dominant language
TypeScript
Stars
133k
Forks
19.9k
Avg merge
18h 46m
Merged PRs (30d)
26

Description

Filed downstream at aviraldua93/gstack-copilot#7 while smoke-testing a fresh install of a port of gstack to GitHub Copilot CLI. The bug exists in canonical upstream gstack — surfaced when the auto-generated /autoplan skill is run on a small plan in a fresh environment.

## Summary

When /autoplan is invoked, the skill mandates a heavyweight pipeline: load 4 sibling SKILL.md files (plan-ceo-review, plan-design-review, plan-eng-review, plan-devex-review) at full depth, dispatch 4 sequential independent Claude subagents, run 4 sequential independent Codex calls, produce 8+ structured deliverables per phase, and present a Final Approval Gate. Pre-gate verification has 30+ checklist items.

This is the right shape for a real review. It is the wrong shape for: (1) smoke-testing a fresh install, (2) quick reviews on small plans, (3) plans that do not trigger all 4 scopes.

There is no documented --smoke-test, --fast, or --tier mode. The only escape is to silently violate the spec.

## Suggested fix

Add an explicit tier selector:
- `/autoplan --tier smoke` — 1 combined subagent, no SKILL.md loads, no structured outputs, final gate only
- `/autoplan --tier standard` — current behavior
- `/autoplan --tier full` — current + dual voices + decision-tree drilldowns

Plus auto-detect: < 200 lines, single branch, no design doc → recommend smoke mode via AskUserQuestion.

## Full evidence and discussion

Cross-posted from aviraldua93/gstack-copilot#7. Happy to PR a patch against utoplan/SKILL.md.tmpl if there is interest.

---
*Filed via GitHub Copilot CLI by @aviraldua93 during stress-testing a downstream port.*

Contributor guide

Open the contributing guide

Research direction

Start with SKILL.md.tmpl and read the current /autoplan pipeline, including its subagent, review, and approval steps. Define how smoke, standard, and full tiers differ, then verify the small-plan auto-detection and recommendation behavior against the issue’s stated criteria. Done means each selector produces the corresponding workflow without silently bypassing the existing specification.

Written by the indexing model from the issue text.

Assessment

Tech stack
github
Domain
cli, tooling
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
52/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.