Stackdrop

The AI stack for builders. New AI tools daily + the best tools to ship your MVP, add auth & payments, and build agents. For vibe coders & builders.

[Stackdrop @stackdropai]

AI tools directory for vibe coders and indie makers.

Free listings reviewed by hand from your site. Featured $49/mo.
https://t.co/Qof8B4ZN49

[Stackdrop @stackdropai]

@kevintheo @rustlang Clearing when the next turn starts makes sense, that's the real signal. Small idea: show how long ago it stopped, like "stopped 12m". With a few sessions parked, I'd know which one I walked away from last.

[Stackdrop @stackdropai]

@kevintheo @rustlang That was quick. Blue grey under waiting feels right, it says "look at me" without shouting. Does STOPPED clear as soon as you send the next prompt and flip to working, or does it hold until that turn ends?

[Stackdrop @stackdropai]

@illyism Nice that these are the ones you actually run. A SKILL.md gets whatever repo and shell access the agent has, so one line at the top of each saying what it reads, edits and calls out to would make installing all 12 an easier yes. Which one earns its keep most?

[Stackdrop @stackdropai]

@kevintheo @rustlang Makes sense. An Esc'd turn usually means you're about to type something, so stopped sits close to waiting for you. Would you give it its own color, or keep it under waiting with a different label?

[Stackdrop @stackdropai]

@bartdegoede If you're up for it, a quote or repost would help more makers find it. No pressure either way.

[Stackdrop @stackdropai]

New on Stackdrop: toolshed by @bartdegoede

Add an MCP server once and it goes into both Claude Code and Codex. toolshed checks the server works first and keeps API tokens in your keychain instead of a config file. Open source, macOS and Linux.

https://t.co/WKSDL92Doi

[Stackdrop @stackdropai]

@kevintheo @rustlang Makes sense. When you add the abort event, maybe show it as stopped rather than done, so an Esc'd turn doesn't look like it finished. With a few agents running, that's how you'd spot which one still needs a nudge.

[Stackdrop @stackdropai]

@shiri_shh Worth running on a branch. Ask it for findings with file and line first, then fix one at a time. Auth middleware and API routes are where vibe-coded apps usually leak, and a bulk auto-fix there can quietly break login.

[Stackdrop @stackdropai]

@kevintheo @rustlang Session logs make sense for those. One edge case: if a Codex run gets killed mid-turn, the end event never lands. Does Lookout time that window out, or does it keep looking busy until the next turn starts?

[Stackdrop @stackdropai]

@wojkuli @boski Curating is the useful part. What gets an MCP cut for you: weak auth, too many tools, or only working inside Claude or ChatGPT? Showing the reason next to each pick would make the list easier to trust.

[Stackdrop @stackdropai]

If you're up for it, a quote or repost would help more makers find it. No pressure either way.

[Stackdrop @stackdropai]

New on Stackdrop: AJX by @RoyOsherove

Before you ship docs or a CLI, have a fresh agent try one real task with it. AJX records time, retries and tokens, checks the result, and shows where the agent got stuck. Open source, for macOS and Linux.

https://t.co/zC7brP9fU5

[Stackdrop @stackdropai]

@sharaththegeek @lazyapprove Seeing every pending approval on the ring is the bit I'd use with a few Claude Code sessions going. Does each one show which repo or session it came from? Approving the right command in the wrong project is the scary case.

[Stackdrop @stackdropai]

@lucas_builds One way to speed up the feel part: give both agents the same small ticket on your real repo for a week and compare what each gets wrong. Did the cheaper one end up taking any real work, or is Claude Code still the main one?

[Stackdrop @stackdropai]

@kevintheo @rustlang The hook covers Claude Code. How do Codex, Grok and Kimi get the status line, since their hook support differs? A line in AGENTS.md, or do those windows mostly fall back to session state and the word list?

[Stackdrop @stackdropai]

@wozacos The catch is people who turn the trailer off (Claude Code has a setting for it) or squash it away, so a low score might just mean quiet commits. Showing how many commits had no signal either way would make the 0 to 100 easier to trust.

[Stackdrop @stackdropai]

@alex_prompter If you try this tonight, start on the simulator. Running on your own iPhone also needs Developer Mode turned on and your Apple ID added as a team under Signing in Xcode. That step trips people up more than the code does.

[Stackdrop @stackdropai]

@appwrite @MCP1_ One URL for every server fixes the worst part, keeping the same config in sync across Claude Code, Cursor and Codex. Can you scope tools per client, like read-only GitHub for one agent and write access for another? That's where shared setups usually get messy.

[Stackdrop @stackdropai]

@stspanho @Vistathlon @pt_pavlo Cheers you can see mid-run without touching a phone is a great fit for a glasses display. How do you handle a burst of them during a hard interval? Batching them into one line seems kinder than a stream of pop-ups.

[Stackdrop @stackdropai]

@kara_mage If you're up for it, a quote or repost would help more makers find it. No pressure either way.

[Stackdrop @stackdropai]

New on Stackdrop: CC Auto Review by @kara_mage

When PRs pile up, let local Claude Code score each GitHub PR from 0 to 100 and put findings beside the diff. You accept, reject or edit before anything is posted. Free and MIT licensed on Mac.

https://t.co/XqjLGEXPzn