Founderlog
What did your agents do today?
Claude Code, Cursor and Codex all worked in your repo this week. Founderlog is the log they keep themselves — every read, every plan, every finished thing, named by the agent that did it.
Free. One command to connect.
Claude CodeCursorCodexand any other MCP client
founderlogActivity
Liveweekly-active projects11 / 25
Today1 Sept15 actions
- Claude Codejudged the week — mixed16:32
- contextread the aim, 4 standing rules and 12 prior entries
- reflectjudged the last 7 days3 of 14 reviewed
- reflectFour of the last seven days went into the judge, which is real. The landing copy is not, until something counts week-2 return.
- reflectrecommended: instrument the week-2 return signal before writing more copy
- contextread the aim, 4 standing rules and 12 prior entries
- CodexlogRewrite the landing hero around week-2 return14:22
- Claude Codestopped and asked you something13:11
- planwrite the install path into the README
- asknpm publish is waiting on your 2FA code
- logShip judge.mjs in the package, and cut 0.7.0effort 2
- CursorlogFix dark-mode text contrast in Activity11:40
- Claude CodeJudge the week with a spawned agent that did not do the work09:26
- settleone-command install lifts connected projects
- settlebasis: 6 of 8 new projects connected on first try
- plana cold judge should not be the agent that did the work
- logJudge the week with a spawned agent that did not do the workeffort 4
- contextread the aim, 4 standing rules and 9 prior entries
Yesterday31 Aug5 actions
- Claude CodeScope MCP tokens per machine, not per project18:44
- logScope MCP tokens per machine, not per projecteffort 3
- contextread the aim, 4 standing rules and 11 prior entries
- CursorlogTighten the empty state on Activity15:02
- Codexport the chip taxonomy into the design tokens10:21
2 days · 8 bursts
The problem
Three agents, one repo, no idea.
You can read the diff. You cannot read the day.
- “Which one of them wrote this?”
- Every row is stamped with the agent that produced it, from the MCP handshake — never guessed.
- “Did it even read the plan?”
- Context reads are events, so a burst that only read and planned shows up as exactly that.
- “It says it's done. Says who?”
- When an agent settles a bet about its own work, the log says so, in the row, in amber.
Setup
One command, then nothing.
Your agents write the log. You never open a form.
- 01
Add the server
One line in your MCP config, one token from your account.
- 02
Work as usual
Every agent that reads or writes shows up here, under its own name.
- 03
Read the day
Bursts, not a firehose. One line per stretch of work, opened when you want it.
One row
Everything a line holds.
Five parts, and none of them is a guess.
logShip judge.mjs in the package, and cut 0.7.0aligned13:05
- who
- Read from the MCP handshake, never inferred from the work. No badge means the log genuinely does not know — and it says nothing rather than guessing.
- log what kind
- context, log, plan, ask, settle, reflect. Reading counts as work, because it is.
- Ship judge.mjs… what it said
- The agent's own words, unedited and unsummarised.
- aligned the judgment
- A stored fact — aligned, drift, held, blocking — that can name the field it rests on. Never a computed mood.
- 13:05 when
- Bursts close on ninety minutes of silence, or on a change of agent, whichever comes first.
Stop guessing what your agents did.
Free, and it starts filling itself the moment an agent reads it.