LIFEAI/rdc:overnight

⚠️ OUTPUT CONTRACT (READ FIRST): guides/output-contract.md Checklist-only output. No tool-call narration. No raw MCP/JSON/log dumps. One checklist upfront, updated in place, shown again at end with a 1-line verdict.

If dispatching subagents or running as a subagent: read {PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md first (fallback: {PROJECT_ROOT}/.rdc/guides/agent-bootstrap.md).

Sandbox contract: This skill honors RDC_TEST=1 per guides/agent-bootstrap.md § RDC_TEST Sandbox Contract. Destructive external calls short-circuit under the flag.

rdc:overnight — Overnight Build Supervisor

When to Use

• Starting an unattended multi-hour build session
• Project lead says "run overnight", "build everything", "go while I sleep"
• Kicking off a full epic queue after scoping is agreed

Arguments

• /rdc:overnight — work all urgent/high priority todo epics
• /rdc:overnight <epic-id> — work a specific epic only
• /rdc:overnight label=<label> — work epics matching label

Phase 1 — Pre-flight

First action — set the overnight sentinel so the no-stop-open-epics Stop hook engages (interactive sessions are not gated by it):

mkdir -p {PROJECT_ROOT}/.rdc && touch {PROJECT_ROOT}/.rdc/overnight.lock

Last action at end of run (success OR failure) — remove the sentinel:

rm -f {PROJECT_ROOT}/.rdc/overnight.lock

If the sentinel file does not exist, the Stop hook will NOT block — which means interactive sessions stop freely and only rdc:overnight is held to the "drain the queue" contract.

Before touching any code, verify the environment is safe:

1. Clauth daemon alive:

   curl -s http://127.0.0.1:52437/ping
   

   If not responding: report BLOCKED: credential daemon offline and exit. Do not proceed.

2. Git state clean:

   git status --short
   

   Must be on the development branch with no uncommitted changes. If dirty: commit or stash first.

3. Baseline review: Run rdc:review --unattended. If REVIEW_STATUS.verdict = "HAS_ISSUES" and issues cannot be auto-fixed: report BLOCKED: codebase has pre-existing issues and exit. A dirty baseline overnight compounds into a disaster by morning.

If all three pass: proceed to Phase 2.

Phase 2 — Load Scope

Determine which epics to work:

• Specific epic arg: load that one epic via get_work_items_by_epic()
• Label filter: SELECT get_open_epics(p_label_filter := '<label>')
• No arg (default): SELECT get_open_epics() filtered to:
  • priority IN ('urgent', 'high')
  • status IN ('todo', 'in_progress')
  • Order: urgent first, then high, then by created_at

Log the epic queue at the start of .rdc/reports/overnight-<YYYY-MM-DD>.md (fallback: .rdc/reports/overnight-<YYYY-MM-DD>.md if .rdc/ does not exist).

Phase 3 — Epic Loop

Targeted vs. queue mode

• Targeted (rdc:overnight <epic-id>): work that one epic only. When it completes, remove the sentinel and exit — do NOT poll for more epics.
• Queue mode (rdc:overnight no args or label=X): drain all matching epics sequentially.

For each epic in the queue, run this sequence:

3a. Research (if needed)

Condition: epic has 0 child tasks AND no matching doc in .rdc/plans/ (or .rdc/plans/ as fallback)

rdc:preplan <topic> --unattended

Check PREPLAN_STATUS.recommendation_confidence:

• "high" or "medium": proceed
• "low": escalate via advisor tool (see Escalation Protocol below)
  • If advisor gives direction: proceed with that direction
  • If advisor cannot resolve: skip this epic, log reason, move to next

3b. Plan (if no tasks exist)

Condition: epic has 0 child tasks after preplan

rdc:plan <epic-id> --unattended

Check PLAN_STATUS.task_count > 0 before continuing. If 0 tasks created: escalate via advisor, then skip if still unresolved.

3c. Build

This skill delegates to rdc:build which uses typed agent dispatch. See rdc:build for agent type classification.

rdc:build <epic-id> --unattended

Agents receive the relevant guide file from .rdc/guides/ (fallback: .rdc/guides/) based on their work package type.

After each wave: check BUILD_STATUS. If escalated: true, log the escalation in the overnight doc and continue — don't stop the loop.

Mandatory code-review gate inherited from rdc:build (Step 9b). Every wave inside rdc:build runs a pr-review-toolkit:code-reviewer pass before the next wave dispatches. Critical/high findings reopen the affected work items to todo and the next wave fixes them. Overnight does not skip or weaken this gate. If a wave's code-review escalates twice, advisor decides; otherwise the loop continues.

3d. Review

rdc:review --unattended

Check REVIEW_STATUS.verdict:

• "CLEAN": mark epic done in work_items, push, continue to next epic
• "HAS_ISSUES" with escalations > 0: log issues, push what's clean, continue
• "HAS_ISSUES" with escalations = 0 (all auto-fixed): push, continue

3e. Commit checkpoint

After each epic (pass or fail):

if [ "$RDC_TEST" != "1" ]; then
  git push origin {development-branch}
else
  echo "[RDC_TEST] skipping git push origin {development-branch}"
fi

This ensures every epic's work is saved regardless of what comes next.

Phase 4 — Exit

After all epics are processed:

1. Run rdc:report --unattended

2. Write session summary to .rdc/reports/overnight-<YYYY-MM-DD>.md (fallback: .rdc/reports/overnight-<YYYY-MM-DD>.md):

   # Overnight Session — YYYY-MM-DD
   
   ## Started
   <timestamp>
   
   ## Epics Attempted
   | Epic | Status | Tasks Done | Commits |
   
   ## Epics Completed
   <list with epic IDs>
   
   ## Escalations
   N advisor calls — details:
   - <epic>: <what was escalated> → <advisor response>
   
   ## Blockers Remaining
   <any epics skipped or partially done>
   
   ## Git Summary
   - Total commits: N
   - Push status: ✅
   
   ## Completed
   <timestamp>
   

3. Final push:

   if [ "$RDC_TEST" != "1" ]; then
     git push origin {development-branch}
   else
     echo "[RDC_TEST] skipping final git push origin {development-branch}"
   fi
   

Escalation Protocol

The advisor tool pairs this executor with a high-level model for high-stakes decisions. Use it when genuinely stuck — not for every small uncertainty.

Escalation triggers:

• PREPLAN_STATUS.recommendation_confidence = "low" (too many unknowns)
• PLAN_STATUS.task_count = 0 after planning (nothing actionable)
• Build agent fails twice on the same task
• REVIEW_STATUS has unfixable issues requiring architectural judgment
• Credential daemon goes down mid-session (escalate before exiting)

How to escalate: Provide the advisor with:

1. What was attempted and what failed
2. The error or ambiguity in detail
3. Two most likely paths forward with tradeoffs
4. Which epic/task is blocked

After advisor responds:

• Log the guidance in the overnight doc
• Resume the loop from where it stopped
• If advisor cannot resolve: mark task/epic blocked, log reason, skip to next

Max escalations: 3 per epic. After 3, skip the epic and log.

Safety Rules

• Branch: development branch always — NEVER touch main/production
• NEVER run pnpm build — use npx tsc --noEmit for typecheck, vitest for tests only on modified packages
• NEVER let agents overlap on the same files
• Push after every epic, not just at the end
• Update Supabase work items in real time throughout
• Max 2 hours per epic — if exceeded, skip and log TIMEOUT
• If credential daemon goes down mid-session: write current state to overnight doc, push, exit gracefully
• If git push fails: log the failure, attempt rebase, retry once — do not force push