Rules · the instructions
A rule crystallizes how a task gets done here, so an agent has everything it needs before it starts: the steps that work, what to never do, the context it would otherwise burn time discovering. Briefed that well, it finishes in fewer steps, on a cheaper model, and gets it right more often.
The instructions: how things are done here.
documentedsharedenforcedsignal-basedNever retry a failed Stripe webhook by hand; use the replay script
reinforced 23×Gate deploys on make migrate-check
holding · 41 runsBump the API and the SDK in the same PR; CI misses version skew
approved · serving 3 teamsSkip e2e on hotfixes
contradicted → retiredYour agents get the relevant rules with every task. Rules that keep working get reinforced; rules that get contradicted are retired. Nothing reaches a team without a manager's approval.
→ before every task, agents fetch the relevant rules and facts every answer logged rules reinforced when they work pages updated when facts change
First run · Claude Opus 4.8 · no memory
≈ $6.40
figures it out: full reasoning, retries, dead ends
the lesson is captured and stored in Firmament
Every run after · DeepSeek v4 Flash + the lesson
≈ $0.70
or Claude Haiku 4.5, or GPT-5.1 Codex Mini: whatever is cheap that quarter
same task · any vendor
11% of the costA briefed agent executes instead of exploring: shorter task horizons, fewer wrong turns, and the same result on a cheaper model. Figure the workflow out once on a frontier model, then run it cheaply, every time after.
A rule captures how a task gets done here: the steps that work, what to never do, the context an agent would otherwise discover on its own. Your agents stop stalling on missing information and stop re-deriving what your team already worked out.
A rule earns its place by working. Each one carries a track record, reinforced every time it holds and retired the moment a real outcome contradicts it. Verified by what happened in production, not by someone re-reading a document.
Agents can propose a rule, but a human approves before it spreads. You decide what becomes the team's standard, and you can see exactly when and why each rule changed.
The rest of the platform