The Runtime Behind the Team
Routines run the loop. GitHub keeps the record. You keep the floor.
If a build dies mid-flight, the record doesn’t.
Six pillars, all shipped
Real Enforcement Mechanisms
Promises are cheap — a durable record, a size-based admission policy, and floors that never move.
The Record Never Forgets
“If a build dies mid-flight, nothing is lost — because nothing was only in memory.”
Every fact lives in GitHub: the issue, its dependencies, the pull request, the review, the merge. Nothing important is held in a process, so an interrupted run costs a re-read rather than a recovery. The next scheduled wake sees exactly what the last one left behind.
- GitHub issues and blocked-by links carry the dependency graph
- An interrupted run is picked up on the next wake
- Cloud agents build in parallel, each on its own branch
- Nothing to reconcile — there is no second copy of the truth
Autonomy Within Bounds
“Small work proceeds. Big work asks first.”
Every repo runs the same size-based admission policy. Small, reversible work moves without a human nudge. Larger work requests your attention through Slack before it starts, and a cap on simultaneous requests keeps the queue from drowning you.
- Sizes through the autonomy threshold proceed unattended
- Larger sizes request approval through native Slack
- A pending-approval cap bounds simultaneous requests
- The refreshed plan and size decide — not a stale label
Some Decisions Are Never Delegated
“Security, production, spend, and the kill switch are never delegated.”
No size threshold changes this. Security surface, production release, material spend, and the kill switch stay human-owned, always — the floor autonomy can never cross, no matter how small the diff looks.
- Security surface decisions stay with you
- Production release stays with you
- Material spend stays with you
- The kill switch stays with you
GitHub Is the System of Record
“Issues to PRs to merge, fully traced.”
Every change starts as an issue and ends as a reviewed, merged pull request. Decisions, findings, and history live where you can search and cite them — not in a chat log that scrolls away.
- Every change traces to an issue
- Review findings and conclusions recorded on the PR
- Merge and closure history stays durable and searchable
- No parallel dispatch, approval, or receipt state to reconcile
Slack Requests. GitHub Remembers.
“Slack requests your attention; GitHub keeps the record.”
Approvals and steering happen in the moment, in Slack — a plan to approve, a PR to review, a question to answer. Slack is a notification surface, not a ledger; the durable record of what happened and why always lives on GitHub.
- Plan approvals and PR review requests arrive in Slack
- One tap approves — GitHub records the outcome
- Slack is never the ticket queue
- Nothing durable is lost if a Slack thread disappears
Nothing Is Learned Twice
“Every decision your team makes is written down, embedded, and searched before the next one.”
Each repo carries a committed brain/ corpus of durable decisions, conventions, and lessons, embedded for semantic retrieval. Agents search it before making a new decision, then write back what they learned — so the same mistake never gets made twice.
- A committed knowledge corpus, per repo, in version control
- Semantic search before new decisions are made
- Durable learnings written back, not lost to a session
- Citations name the page a decision came from
The Runtime Runs the Team
This is the machinery. See who it's running — portfolio agent, resident repo agents, merge-denied builders, and an independent reviewer.
Start Shipping While You Sleep
Six pillars. One runtime. You stay in control.