status: active
timestamp: 2026-06-23
tags: [rule, agents-md, knowledge, context-management, discipline]
2025 agents.md discipline — tight context, knowledge bundle, don't drift to 2026 yolo
User locked 2025 mindset 2026-06-23: AGENTS.md stays short sharp, knowledge/ is the brain, every concept gets a file. Reject the 'just let\ agents figure it out from context' 2026 yolo for non-toy projects. For oriz family\ this means: AGENTS.md ≤200 lines pointing at README.md and knowledge/, never\ inline rules in AGENTS.md if a knowledge file exists, prune stale knowledge weekly,\ treat context as a precious limited resource.
Rule: 2025 agents.md discipline
What
For this repo and the oriz family:
AGENTS.mdstays =200 lines. It POINTS atREADME.md(canonical entry) andknowledge/index.md(canonical brain). It does NOT inline rules that have their own knowledge file.- Every multi-decision concept gets its own file in
knowledge/rules/<slug>.md,knowledge/decisions/<slug>.md, orknowledge/runbooks/<slug>-YYYY-MM-DD.md. - Each markdown front-matter follows the OKF v0.1 spec (
knowledge/_okf.md). - Context is precious. Adding a rule means deleting another, or splitting an existing one. The total agent-context payload should be measured, not assumed.
- Stale knowledge gets PRUNED, not preserved out of politeness. If a decision is superseded, mark it
status: supersededand link the successor.
Why
User pasted the 2025-vs-2026 mindset debate 2026-06-23 and locked the 2025 mindset for this project. Direct quote: “I think it’s more nuanced than this… at least for now, I suggest the same for you… for larger projects, I like to be all over it.”
This project is NOT a toy. It is the oriz family — 60+ submodules, paying-user infra, family-level cohesion via shared chrome and design tokens. The 2026 “let it hang out” approach is appropriate for a first-person shooter built in a weekend. It is not appropriate here.
How to apply
- Before adding to
AGENTS.md, ask: does this belong inknowledge/rules/,knowledge/decisions/, orknowledge/runbooks/? If yes, write it there + add a one-line pointer inknowledge/index.md. - When making a decision, write the decision file FIRST, then code from it. Don’t write code-then-doc.
- Treat
AGENTS.mdupdates as expensive. Every line there is loaded into every agent’s context on every invocation across 60 repos. - Run the agents-md-sync skill periodically to dedupe overlapping rules across
claude.md/agents.md/gemini.mdif the project ever adds those formats. - When pruning: a knowledge file marked
supersededfor >30 days can be deleted entirely if its successor has absorbed the relevant content. Git history is the durable backup.
How NOT to apply
- DON’T let
AGENTS.mdgrow past 200 lines. If it does, it’s a smell — extract sections toknowledge/. - DON’T duplicate rules across
AGENTS.mdandknowledge/rules/*.md. The knowledge file is the source of truth;AGENTS.mdpoints at it. - DON’T preserve stale knowledge “just in case”. The “just in case” is git history.
- DON’T add a
knowledge/file for one-off decisions that won’t recur. The bar is “would a future agent need this to make a similar decision again.”
Cross-refs
- README.md is the canonical entry — see README.md.
- The knowledge brain index — see knowledge/index.md.
- Format spec for files in
knowledge/— see knowledge/_okf.md. - Companion rule on filing every locked decision — see [[self-update-on-every-decision]].
- Companion rule on confirming knowledge deltas — see [[confirm-knowledge-deltas]].