Skip to content
~/agentops
DWG. 12-FACTOR AGENTOPS — THE DOCTRINEREV 2026.06 · SHEET 01 / 07
// 12-FACTOR AGENTOPS

You're running production
on vibes.

Agents generate code with no record of what was tried, no gate it had to pass, no proof it works. 12-Factor AgentOps is the operating model that closes that gap. Twelve factors that make agent-built software something a team can stand behind, in the tradition of the Twelve-Factor App.

Read the 12 factors →Run it in your agent →GitHub ↗
FIG. 01 — THE FOUR COMPOUNDING LAYERS
LAYER 1
BOOKKEEPING
Record attempts, decisions, verdicts in .agents/.
Never wonder what the agent already tried.
LAYER 2
CONTEXT COMPILER
Compile the right slice into the next run.
No re-explaining your codebase each session.
LAYER 3
VALIDATION GATES
Fresh-context review before plans & code ship.
Catch broken work before it hits your branch.
LAYER 4
KNOWLEDGE FLYWHEEL
Promote learnings into durable constraints.
Solved once, stays solved.
├ cold start↺ THE CORPUS COMPOUNDS — NEXT SESSION STARTS LOADEDcompounding ┤
// the 12 factorsall 12 →

The operating rules the site is named for: twelve, in four tiers.

IFOUNDATION
Context
II
Track
III
Scope
IVFLOW
Research
V
Validate
VI
Lock
VIIKNOWLEDGE
Extract
VIII
Compound
IX
Measure
XSCALE
Isolate
XI
Supervise
XII
Harvest
// see it work

A skill loads the corpus before writing a line of code; a mixed-model council reviews before it ships.

> /research add rate limiting to /login
// loading context from .agents/ …
3 prior auth decisions cited · 2 planning rules · 1 learning
plan: token bucket, 5/min per IP, Redis-backed, jittered
// recorded → .agents/runs/2026-05-08-rate-limit/research.md
> /council --mixed validate this PR
// evidence packet sealed → 6 judges across Claude Code + Codex
claude · WARN rate limiting missing on /login
codex · WARN token-bucket refill lacks jitter
consensus: WARN — fix before shipping
// the bet

THE WAGER — Vendors will ship managed memory, review councils, and overnight learning loops natively; they will lock them to their runtime. Your corpus stays in .agents/ in your repo, runs on whichever harness you already pay for, and is portable across whichever frontier model wins next quarter.

Good architecture principles outlive every tool that implements them.
see the proof →
// the three gaps it closes

The failure modes that make agent work unreliable, each closed by a named surface.

G1JUDGMENTPressure-test plans before code.closed by /pre-mortem · /vibe · /council
G2DURABLE LEARNINGSolved problems stay solved.closed by /retro · /forge · ao lookup
G3LOOP CLOSUREShipped work informs the next session.closed by /post-mortem · finding compiler · /evolve
// SHEET INDEXTHE SET, IN ORDER ↓
02/factorsthe twelve, in four tiers→03/cdlcthe context development lifecycle→04/comparisonshow it stacks up against alternatives→05/journeydoctrine evolution as a devlog arc→06/skillsthe catalog, enforced in your runtime→07/installrun the loop in your harness→
// the standard for software you can trust an agent to build.
the corpus stays yours_