# agentsop.ai/sop

> Standard operating procedures in three layers: self-check, how-to, catalog.

Repository: [repos] sop in ~/.config/agent-sop/config.toml, else sources.sop in the org repository's org.json.

## 1. Self-check

- `AGENTS.md` (sop repository): Read before acting.
- `docs/reference/agent-conventions.md` (sop repository): PR-first, validate before push, untrusted input, progressive disclosure.

## 2. How-to

- `docs/sop/README.md` (sop repository): The SOP index. Inherited by default, varied only by a recorded delta.
- `docs/sop/issue-lifecycle.md` (sop repository): Every change traces to an issue.
- `docs/sop/branch-pr-review.md` (sop repository): One objective per PR and the review bar.
- `docs/decisions/README.md` (sop repository): Cross-repository decisions (the ENG series) with status.

## 3. Catalog

- `skills/README.md` (sop repository): Shared skills. Load one only when a procedure names it.
- `org.json` (org repository): The capabilities table: the tools pinned for the organization, each with its entry file.

## More

- [llms-full.txt](https://agentsop.ai/sop/llms-full.txt): The remaining SOPs, the guide index, and how to choose between an SOP, a decision, and a guide.