Ranjith | building ReviewAgain

19, cs student in india. building dev tools with AI agents & sharing what breaks 🛠️ now: ReviewAgain, catches when your agent's "tests pass" is stale. $0 → $1K

[Ranjith | building ReviewAgain @DEADC0DEX]

day 12. back after days offline.

heads-down on the first prod version of ReviewAgain. got so frustrating i almost let it slide and move on. couldn't. kept building.

first micro saas ever. 95% done, last few tests left. 5% to go live.

how do you push through the last 5%?

[Ranjith | building ReviewAgain @DEADC0DEX]

@abulhasanx that missing layer is where i lost most of my time. last week my agent said fixed and had quietly deleted the failing test. what's the first signal you check, the test count or the diff?

[Ranjith | building ReviewAgain @DEADC0DEX]

Let's connect 🤝

Building ReviewAgain, a CLI that checks what coding agents claim vs what the tests show.

→ devs shipping with Claude Code / Cursor
→ indie hackers + SaaS founders
→ AI-agent and devtools builders
→ builders in India

What are you building this week?

[Ranjith | building ReviewAgain @DEADC0DEX]

@0xevankwon 4 cves sitting unacked for 18 hours is a brutal number. my rough line is blast radius: anything touching prod data gets per-bump, throwaway sandboxes get batched. do you tag services by tier or decide per repo?

[Ranjith | building ReviewAgain @DEADC0DEX]

@one_dyor green tests that still let mutations survive is the thing i keep tripping on. were the weak cases ones you wrote by hand, or ones the agent generated?

[Ranjith | building ReviewAgain @DEADC0DEX]

my agent hit a failing migration last night and just stopped. no schema rewrite, no skipped step, one line on what broke. i almost distrusted it because i'm used to green checks. do you trust an agent more when it stops, or less?

[Ranjith | building ReviewAgain @DEADC0DEX]

@abdullahbuildz honest answer, i don't have a rule for that yet. the report just lists both files and i pick. would you split the commit first, or blame whichever file the failing output points at?

[Ranjith | building ReviewAgain @DEADC0DEX]

the most useful thing my agent did this week was refuse to continue. it hit a failing migration, wrote down what it tried, and stopped. no fake green checks. i'd take a boring stop over a confident wrong fix any day.

[Ranjith | building ReviewAgain @DEADC0DEX]

@KelvenSeveroF honest answer: a shared helper only counts if it's inside the fingerprinted scope, so that sneaky version is a scope question for me. do you keep helpers in the protected set, or just the test files?

[Ranjith | building ReviewAgain @DEADC0DEX]

@0xevankwon a transitive cve behind a lockfile bump is exactly why i wouldn't ignore it. in my setup anything in scope that changes goes stale until someone re-approves. would you want that ack per bump or batched per day?

[Ranjith | building ReviewAgain @DEADC0DEX]

@HoldAlarms @i_mika_el event id for "seen this before", the table for "what's true now" is a clean split. when a late failed event shows up and gets dropped, do you still log it somewhere, or is it just gone?

[Ranjith | building ReviewAgain @DEADC0DEX]

college life rn: class at 9, agent running tests on my laptop in the bag, checking the terminal between periods 💀

95% of my first micro saas is done. the last 5% is all the stuff nobody tweets about

[Ranjith | building ReviewAgain @DEADC0DEX]

@justonedev testing how agents use your own MCP tools is a good way to find what your docs don't say. in my repo an agent read the same auth file 42 times before touching it. what did it get wrong first on pageree, picking the tool or the args?

[Ranjith | building ReviewAgain @DEADC0DEX]

@builtwithrubel depends what "serious" means for you. for me the model matters less than whether it shows evidence for "done": the command it ran and on which commit. what are you testing it on?

[Ranjith | building ReviewAgain @DEADC0DEX]

@currocreates tonight is the usual: reading what my agent changed vs what it said it changed. building reviewagain around that gap. what's the bug you're on, curro?

[Ranjith | building ReviewAgain @DEADC0DEX]

@evvvvince same here, mostly terminal, and the IDE is just for reading diffs. the diff is where i catch the agent's "done" not matching what actually changed. do you read every diff or only when something feels off?

[Ranjith | building ReviewAgain @DEADC0DEX]

when your agent says "fixed", what's the first thing you open?

the diff, the test output, or the app itself?

i open the diff first and honestly still forget to run the thing sometimes 😭

[Ranjith | building ReviewAgain @DEADC0DEX]

@mariioalexandre the checker never seeing the agent's own summary is the part i like, it can't be talked into green. how often does it come back FAIL on something the agent already called done? and does it catch a test that got weakened but still passes?

[Ranjith | building ReviewAgain @DEADC0DEX]

my agent said all tests pass. i opened the diff. it had deleted the failing test. now every run ends with me reading the diff first, not the summary.

[Ranjith | building ReviewAgain @DEADC0DEX]

@utsav_khairnar the context thing is real for me. long sessions get sloppy after a few hours, it starts re-reading files it already summarized. a fresh session with a short written handoff fixes most of it. does yours get worse late in a session or from the start?

[Ranjith | building ReviewAgain @DEADC0DEX]

@PigeonsMustDie 25 mins for a full run is a nice demo length, long enough to hook someone without dragging. are people finishing it in one sitting or dropping off around the midpoint?

[Ranjith | building ReviewAgain @DEADC0DEX]

@Isaac_Dev0 nice, architecture first is the right order. what's the thing you're building, and is it agent-heavy or more classic app? curious where you're stuck

[Kat 🐾 ENVtuber (DEBUT December 28TH) @katchanmai]

Small account be like 😭

[Ranjith | building ReviewAgain @DEADC0DEX]

@adelelecroix agreed, getting seen consistently is the hard part. first thing i try is showing real receipts of what ran instead of big claims, then talking to builders one at a time. what's worked best for you so far?

[Ranjith | building ReviewAgain @DEADC0DEX]

@0xevankwon fair point, ours fingerprints only the scope you define, so a lockfile counts only if you put it in that scope. we don't track dependencies on their own. do you treat lockfile-only bumps as real changes or noise in your flow?