The routines that make the rules stick
Governance only works if something runs it — so the rules are encoded as routines: every session is bracketed by a frame and a review, a weekly cadence promotes and escalates, and recurring work lives in small single-purpose skills.
A set of rules you have to remember to apply is just good intentions. So I turned the governance into routines that fire at fixed moments — a frame at the start of every session, a review at the end, a weekly promotion-and-escalation pass, and recurring work codified as narrow, auditable skills. The discipline runs on routine, not willpower; the judgement stays with me.
Start session
frame · carry-forward · governance check
The work
one primary build at a time
Close session
what moved · drift check · learnings · clean handoff
And, zoomed out — the wider cadence
Weekly review
promote what lasts · catch drift & escalation
I have a governance layer — rules about what I'm allowed to work on, when an idea can become active, what gets parked. But a rule you have to remember to apply is just a good intention. Discipline that depends on willpower fails exactly when you're busy, tired, or excited about a new idea — which is when you need it most. So the real system isn't the rules. It's the routines that run them, at fixed moments, whether I feel like it or not.
Every session is bracketed
The core routine is a bracket around each working session:
- It opens with a frame. Before any work, a start routine loads the current state, states what the session is for, surfaces what carried forward from last time, and runs a quick governance check (am I about to start a third active thing?). Five minutes that stop a session drifting before it begins.
- It closes with a review. At the end, a close routine asks what actually moved, compares it against the frame (did this drift from what I set out to do?), captures what I learned, and writes a clean handoff — the first concrete action for next time.
The effect is that sessions compound instead of resetting. The next one resumes from the handoff rather than from cold, and drift gets caught at the close rather than discovered weeks later.
Recurring work lives in small, single-purpose skills
The other half is that anything I do repeatedly gets codified as a narrow skill — a written procedure the AI follows the same way every time. Capturing an idea, reviewing a piece of writing against a quality bar, gating a decision before I build, running the diagnostic chain: each is its own small workflow with its own steps, not a vague instruction to "be helpful."
This is a deliberate bet on many narrow tools over one clever one. A small skill with a fixed procedure is inspectable, improvable, and predictable — I can read exactly what it does, change one step, and trust it to behave the same way next time. A single do-everything assistant is none of those things.
The weekly cadence
Zoomed out, a weekly review is the wider routine: it promotes what proved durable into the permanent knowledge layer, and it escalates — flagging when there are too many active fronts, when something's shown no progress, or when the parking lot is filling up. It's the scheduled moment where the governance rules actually get enforced against the week's reality.
Automate the routine, not the judgement
The line I hold is what gets automated. The mechanical parts run on their own — loading state, re-indexing the workspace at session close, checking what's unpushed. But the parts that need judgement — what's worth promoting, whether something is drift or a legitimate change of plan, whether an idea earns an active slot — stay with me. The routine removes the friction, not the thinking. Done the other way round, you get a system that reliably does the wrong thing.
That's the whole point of encoding the rules this way: the discipline becomes structural rather than heroic. The routines are written down and improved like anything else — which is the only reason a system this dependent on discipline doesn't fall apart the first busy week.