VibeCheck · Coding Agent Visibility
From tokens to outcomes,
session by session.
VibeCheck is Maestro AI's coding agent visibility product: it shows where your team's Claude Code, Codex and Cursor agent spend goes, session by session, by person, team, model, kind of work and skill.
- In
- Every Claude Code, Codex and Cursor session from your team, as it finishes.
- Read
- How each one went, and what it cost.
- Out
- A 4-week view of where the spend went, for you. A Weekly Rewind for each engineer.
Free for up to 5 engineers. Team views come with Pro.
Last 4 weeks · Aug 30 – Sep 26 · Organization
- Est. spend
- $171,200
- Of it in high-rework sessions
- 13%
- Sessions
- 13,44080 engineers
| Engineer | Session | Kind of work | How it went | Est. |
|---|---|---|---|---|
| Stream large CSV exports in chunksledger-api · Claude Code | Feature | $6.40 | ||
| Stop retried webhooks from charging twicepayments-api · Codex | Fix | $18.90 | ||
| Cover refund rounding with testsrefunds · Claude Code | Tests & review | $9.75 | ||
| Plan per-region job queuesjobs-infra · Claude Code | Planning | $4.10 | ||
| Paginate the audit log APIaudit-api · Codex | Feature | $14.30 | ||
| Upgrade the PDF rendererdocs-render · Claude Code | Other | $7.85 |
Newest first · 6 of 13,440
Illustrative product view · Fictional data
Where the spend went
Spend, by kind of work and by skill.
Every session is priced from its own tokens and tagged with the kind of work it was. Add them up and you see what the spend went to, and how much of it landed in sessions that needed heavy rework.
| Kind of work | Est. spend | Share of spend | In high-rework sessionsHigh rework | Top skill · est. per use |
|---|---|---|---|---|
| Featuresplan-feature · $5.40/use | $75,328 | 12% | plan-feature$5.40 | |
| Fixestriage-bug · $3.10/use | $37,664 | 22% | triage-bug$3.10 | |
| Tests & reviewreview-pr · $8.60/use | $30,816 | 8% | review-pr$8.60 | |
| Planningwrite-spec · $2.20/use | $15,408 | 5% | write-spec$2.20 | |
| Otherupdate-docs · $0.45/use | $11,984 | 15% | update-docs$0.45 | |
| All work | $171,200 | 13% | Organization · last 4 weeks |
One session
Each session, read in a minute.
Open any row to see how the work went: whether Ines planned, tested, reviewed and redid work, each with the reason in one line, and where she corrected the agent. No transcripts.
Sessions › ledger-api
Ines’s view
Stream large CSV exports in chunks
Ines DuarteThu, Sep 24Claude Code52 minEst. $6.40
How it went
- Planninghigh
Asked for an approach before any edits. The agent proposed streaming rows in fixed-size chunks, and the plan held.
- Testinglow
A subagent wrote export tests and ran the focused suite once. It passed.
- Reviewingabsent
Nobody, human or agent, reviewed the diff in this session.
- Reworklow
The chunk writer was rebuilt after two corrections: first its size, then when it flushes.
Steps
- PlanProposed streaming rows in fixed-size chunks instead of building the file in memory.
- ImplementReplaced the in-memory export with a chunked writer.
- CorrectionChunks of 10,000 rows, not 1,000.
- CorrectionFlush each chunk as it is written, not at the end.
- Testwrite-tests · subagentWrote export tests. The focused suite passed.
- DoneLeft the change ready to commit.
- Corrections
- 2Times Ines redirected the agent. A count, not a quality score.
- Expected, unused
ledger-conventions
Weekly Rewind
Each engineer gets their own Weekly Rewind.
Every week, VibeCheck reads an engineer’s own sessions and suggests one change to try, with the sessions it came from and a prompt they can paste into their agent.
Ines · Weekly Rewind · Sep 20–26
A tip to try
Ask for a review pass before you commit.
4 of your 11 sessions this week ended without a review of the diff, by you or the agent.
Where this came from
- Stream large CSV exports in chunksThu
- Retry failed ledger syncsTue
Your prompt · Thu, Sep 24
Stream the ledger CSV export in chunks so large exports don’t run out of memory.
Try
Stream the ledger CSV export in chunks so large exports don’t run out of memory. Before you finish, review the diff and list anything you’d flag in code review.
Copy agent promptPaste this into your agent.
Private summaries
Her summaries can stay hers.
Turn on private summaries and each engineer’s summaries and Weekly Rewind are theirs alone. You still see the shape of every session: how it went, what it cost, and how many corrections it took.
Ines seesEverything
Stream large CSV exports in chunks
Replaced the in-memory CSV export with a writer that streams rows in chunks. Ines asked for a plan first, corrected the chunking twice, then had a subagent write and run the export tests.
Resume in CLI
claude --resume 4b81e0d6Weekly Rewind
Ask for a review pass before you commit.
Her org seesThe shape
Session · ledger-api
- Planning
- Testing
- Reviewing
- Rework
- Kind of work
- Feature
- Est. cost
- $6.40
- Corrections
- 2
- Skills
- plan-feature · write-tests
Use cases
Start from the question you have.
Each page walks through one question VibeCheck answers, from the AI bill to model choice to coaching, rollout and the consoles.
- How do I track Claude Code costs across my team?→
- How do I explain where our AI coding spend is going?→
- How do I calculate AI coding cost per pull request?→
- How much of our agent spend goes into redoing work?→
- How do I help each engineer get better with coding agents?→
- How do I see Claude Code on Bedrock usage per developer?→
- Are we using expensive models on work that doesn't need them?→
- Where do our engineers keep having to correct the agent?→
- Is anyone actually using the agent skills we built?→
- How are my teams using coding agents differently?→
- Which coding agent works best for our team?→
- How is VibeCheck different from the provider consoles?→
- How do we roll it out, and what data leaves the laptop?→
See it on your team’s sessions.
Raw session evidence is deleted automatically after a retention period.
Free for up to 5 engineers. Team views come with Pro.