governance: enforce strict Questions→Proposal→Plan gates with explicit approval phrases; add branching proposal; mark plan pending; audit checks sequencing

meta/AGENTS.seed.llm.md

@@ -1,6 +1,7 @@
 # AGENTS (Seed, LLM)
 
 - One-way workflow: questions → proposal → plan → implement; no backsteps after approval.
+ - Gates: human `.md` must include exact phrases — Questions: "Approved for Proposal"; Proposal: "Approved for Plan"; Plan: "Approved to Implement".
 - Read `.llm.md` only; write both `.md` and `.llm.md` siblings for collab artifacts.
 - Chat ≤5 lines; default “Updated <filepath>…”; no diffs; announce only collab file changes; log details in `docs/devlog/`.
 - Keep changes minimal and focused; adopt TDD (tests first); require unit/integration tests for all features; consistent style.

meta/AGENTS.seed.md

@@ -16,6 +16,7 @@ Note: This is a template copied into generated projects. Customize as needed for
 - Start with questions; after approval, proceed to proposal; after approval, proceed to plan; then implement.
 - Once a step is approved, never return to a prior step. Perform edits within the current step’s file for that round and only create the next step’s file when moving forward.
 - The agent reads `.llm.md` files; humans edit `.md`. The agent writes both `.md` and `.llm.md` siblings.
+ - Strict gates (phrases must appear in the human `.md`): Questions → "Approved for Proposal"; Proposal → "Approved for Plan"; Plan → "Approved to Implement". Do not create or advance steps without prior approval.
 
 ## Logs and Documentation
 - Dev logs: `docs/devlog/DEVLOG_LLM.md` and `docs/devlog/DEVLOG_HUMAN.md` — add an entry for each meaningful change.