| name | despaghettify-autonomous-agent |
| description | Routes agents to the Despaghettify autonomous loop — Step 0 backlog drain uses slice → check --out → autonomous-advance backlog-implement (--check-json repo-relative); only after the DS row is closed in the input list, backlog-solve; then main check→solve until trigger clears or hard stop. |
Despaghettify autonomous agent (router)
Do not duplicate the loop or policy here. The only specification for the check → solve → check macro loop, success conditions, error scope, and anti-stall rules is:
despaghettify/spaghetti-autonomous-agent-task.md
Read that file in full and execute it. Each check pass must follow despaghettify/spaghetti-check-task.md (after reading numeric policy in despaghettify/spaghetti-setup.md). Each solve pass must follow despaghettify/spaghetti-solve-task.md for one DS-* at a time.
Machine guards (must): autonomous-init at session start (repo root: python -m despaghettify.tools …). Step 0 — while a backlog DS row is open: after each implementation slice, run hub check … --out <file>, then autonomous-advance --kind backlog-implement --ds DS-0xx --check-json <path> where <path> is relative to the repository root. Only after the DS goal is met and the row is closed in despaghettification_implementation_input.md: autonomous-advance --kind backlog-solve --ds DS-0xx (that advance exit 2 if the row is still open — HARD_STOP). For all other macro steps, autonomous-advance as in the task doc (exit 2 → HARD_STOP). Run autonomous-verify between waves as in the task doc. Regenerate despaghettify/spaghetti-setup.json with setup-sync after edits to spaghetti-setup.md (JSON is derived only, not hand-maintained).
Optional CLI: python -m despaghettify.tools check | open-ds | solve-preflight --ds DS-0xx — see ../references/CLI.md.
