build

Prompt file imported from edwardlthompson/MultiAppShare- (.cursor/commands/build.md). Copyright stays with the author.

Autonomous sprint build (end-to-end, automate HUMAN/ADB first)

Execute the BUILD_PLAN without asking the user questions, presenting options, or waiting for approval. Pick the single best path internally (### Critique + ### Parallelization in your reasoning only), then implement it.

Goal: Complete as much of BUILD_PLAN as possible in one run. Run all [AGENT]/[AUTO] and Parallel work first, then attempt the grouped Human & device (after automation) section. Chain sprints until all actionable rows are done, 3-strike failure, or environment block.

Rules

• Never stop to ask "which approach?" — decide and execute.
• Never stop for plan approval — /build is self-approving.
• Never halt on [HUMAN] or [ADB] labels — automate first; backlog only on failure.
• Never stop after a single feature row if the active sprint still has open work.
• Do stop only when: (a) 3-strike gate failure, (b) environment block (exit 2) after autofix, or (c) all actionable rows complete (all_sprints_agent_auto_complete).
• On gate failure: run @.cursor/commands/fix.md autonomously (up to 3 strikes) — do not suggest /fix and wait.

Step 0 — Load sprint state

Use --lane auto (not child): on this template it selects Ongoing Maintenance; on child repos it still walks the Child Repo Playbook.

python3 scripts/agent-run.py build-sprint-status --json --lane auto

auto uses the child playbook on product repos and the Template Maintainer board on this template. Pass --lane child only when you mean Sprint 0+.

Write .cursor-session-state.json fields: active_sprint, build_plan_lane, autonomous_mode: true.

If all_sprints_agent_auto_complete: true → print summary (include HUMAN_BACKLOG.md path if items exist) and exit.

Step 1 — Sprint execution loop

Repeat until sprint_agent_auto_complete:

1a. Read status

python3 scripts/agent-run.py build-sprint-status --json --lane auto

• If next_row is null and sprint_agent_auto_complete → go to Step 2 (sprint wrap-up).

1b. Execute next_row

1c. Gate autofix (every AGENT step)

python3 scripts/agent-run.py watch-agent-gates --once --autofix --scope auto

--scope auto gates only dirty stacks (git vs HEAD: examples/{stack}/, or preamble-only for docs/changelog/commands). Shared scripts/ / schemas/ / modules/ still run the full multi-stack gate. Print the gate scope: line. Override with --scope full or FEATURE_GATE_SCOPE=full.

Exit 1 → fix in scope and re-run (3-strike max). Exit 2 after 3 strikes → halt with evidence; do not ask user.

Skip gates for successful automate_human/automate_adb no-op/informational steps unless the automation script ran product smoke or init.

1d. Loop

Gate lock: do not execute another next_row until step 1c watch-agent-gates exited 0. Never skip 1c to chain a second feature.

Re-run build-sprint-status.sh --json and continue 1a.

Step 2 — Sprint wrap-up

When sprint_agent_auto_complete for current sprint:

1. @.cursor/commands/gates.md — full local validation
2. @.cursor/commands/cleanup.md — archive ✅ rows (including auto-completed HUMAN/ADB); backlog items stay open on board
3. Print brief summary: sprint name, rows completed, rows automated, rows backlogged (HUMAN_BACKLOG.md), and pointer to grouped Human & device (after automation) section for manual follow-up

Step 3 — Chain to next sprint

Re-run python3 scripts/agent-run.py build-sprint-status --json --lane auto.

• If next_row exists → go to Step 1 immediately (no user pause).
• If all_sprints_agent_auto_complete → print final summary: all actionable BUILD_PLAN work complete; list backlogged_human_adb and HUMAN_BACKLOG.md path.

Progress logging

Log one line per completed row (owner + task + automated|implemented). Do not dump full plans between rows. User sees sprint summaries only at wrap-up and final exit.

Begin now. Do not ask the user anything until the loop exits.