A day with Faundr

This page follows a typical session, from the moment you open your AI agent to the end of the day, and shows what Faundr does at each step. You don't need to memorize anything. Most of it happens on its own.

When you open your agent

As soon as the session starts, Faundr hands the agent a summary of the project, before you've typed a word:

• where the last session left off: the requests, commits, completed tasks and the handoff note;
• the team's rules, which the agent should always follow;
• decisions already made, so it doesn't propose something that was already settled;
• the feature in progress and what's left in it;
• open concerns and anything new in security, errors, quality and tests;
• team notes, if someone left one in the dashboard;
• your team's active workflows, with when to use each one.

On the first prompt of the session, Faundr also runs the non-AI checks in the background: security, quality, design, Stack, environment variables and the test inventory. It also rereads your skills, so a new one shows up in the dashboard without you doing anything. None of this slows down the agent's reply.

When you start something new

For a new feature, just ask for it. The agent plans it with you first (/faundr:planning). To create it as this folder's current feature right away, use /faundr:feature:

/faundr:feature Login screen — sign in with email and password

Before starting, the agent sizes up the request and looks through the project, so it only asks what the code can't answer. Then it asks a few questions about what you see and do on screen (never about technology), each with options and a recommendation, and shows you a summary to approve. Your choices become decisions, and the "done when" becomes the feature's definition of done. Then it breaks the work into tasks, and everything shows up in Features.

You can also just ask for what you want. When the agent starts something with several steps, it records the feature and tasks on its own. A plan approved in the agent's plan mode also becomes a feature with a checklist.

If the work is large and happens in phases, the agent records the next milestones from the start. They show up under Features → Next milestones, and Faundr brings them up when the current milestone wraps.

If a request matches one of your team's active workflows, the agent gets the steps and follows them in order while your team watches live on the canvas. If a request touches the topic of one of your skills, the agent gets a reminder to use it. See Workflows.

While the agent works

Every change goes through a few safeguards without you having to ask:

• Before the first change of the session, Faundr saves a checkpoint of your files. It saves another one every 30 minutes of edits.
• Before editing a file, the agent gets the rules and decisions tied to that file and the list of what depends on it, so it doesn't break another part without noticing.
• If it tries to write a secret key into the code (Stripe, Supabase, AWS, OpenAI…), the edit is blocked and it's told to use .env instead.
• If it tries a dangerous command (deleting folders, git reset --hard, DROP TABLE, touching the production database), the agent stops and asks you to confirm. Anything that can't be undone at all is blocked outright.
• If it tries to install a package that doesn't exist, or one that's suspiciously new, Faundr blocks it or asks you.
• If a command fails, the error becomes an item in Errors. When the same command passes again, the error closes on its own.
• If it goes in circles (the same error 3 times, or the same file changed over and over without a fix), Faundr asks it to stop, explain what's going on and consider rolling back to an earlier checkpoint.

The details are in Protection.

At the end of every response

When the agent finishes a response that changed files, Faundr checks what's still pending. If something's missing, it asks the agent to do it before wrapping up:

1. Prove it works: run the build, the tests or the type check, and report the result.
2. Leave a handoff note in long sessions: 1 to 3 sentences on what's half done and what comes next.
3. Review the screens it changed, if the project has a design.md.
4. Check the definition of done when the feature's checklist is complete.
5. Announce the next milestone and ask whether to continue.
6. Update the Stack or the Overview if they're out of date.
7. Record the work if the session changed several files and nothing was logged in Faundr.
8. Send a connected skill's results to the dashboard if one ran during the response (one of your skills that finds security, test, quality or error problems). What it found shows up in the right section, tagged "via /name".

You can save the quick design review and the Stack and Overview updates for when you ask, with economy mode (/faundr:setup offers it). Proof, the handoff note and security always stay on.

That's why the agent's reply sometimes keeps going after it looked finished, with a line like "I ran the build and it passed." That's Faundr doing its job.

In the background, Faundr also rebuilds the code map, rechecks security, quality and design on the changed files, and adds up the tokens used.

When something goes wrong

At the end of the day

• Before you close, run /faundr:handoff if you want to write the handoff note in your own words. In long sessions, the agent is already reminded to write one.
• Commit what's done. Tasks marked done with uncommitted changes show up as uncommitted in the dashboard until the next commit.
• Open the dashboard to review the day. Activity → Where you left off shows the last session and what's still to do.

The next day, your new session starts out knowing all of this.

Every week

• Activity → Reports covers the last 7 days in plain language: what moved forward, what shipped, the problems and the cost. It can go out automatically every Monday morning to Slack or Discord.
• Insights → Cost shows what your sessions cost, by feature and by 5-hour window.
• Before opening the app to the public, run /faundr:launch and see Launch and privacy.