Daily Loop Runner

Advance one project by one meaningful step per loop. State read before execution, structured writeback after.

Input

Required:

• - project_card — full project card with current state, goals, blockers
• INLINECODE1 — most recent weekly review notes
• INLINECODE2 — list of recent daily loop logs (last 3-5)
• INLINECODE3 — unresolved questions from previous loops

Optional:

• - forced_bottleneck — override automatic bottleneck selection

Output Schema

CODEBLOCK0

Hard Rules

1. 1. One project per run. Do not split focus.
2. One bottleneck per run. Pick the most critical blocker.
3. One main action per run. One meaningful step, not a sprint.
4. No execution without state read. Always read projectcard and recent logs first.
5. No successful completion without nextaction. Every loop must feed into the next.

Loop Phases

Phase 1: State Read

• - Read projectcard fully
• Read recentdailylogs
• Note openquestions from previous runs
• Identify current project phase and milestone

Phase 2: Bottleneck Selection

• - Pick the single most critical bottleneck
• If forced_bottleneck provided, use it
• If nothing is blocking, advance the primary goal

Phase 3: Task Input Construction

• - Build a focused task_input for the selected agent
• Include: what to do, why it matters, what success looks like
• Exclude: everything else

Phase 4: Execution

• - Dispatch task to selected agent
• Wait for execution_summary and findings

Phase 5: Writeback

• - Update projectcard with projectcardupdates
• Write executionsummary + findings to daily log
• Populate nextaction for tomorrow's loop
• Set safeto_proceed = true

Failure Handling

If state is incomplete (missing project_card, no recent logs, unclear objective):

• - Stop the run immediately
• Set INLINECODE5
• Request cleanup or missing-state repair
• Do not attempt execution with partial state

State Machine

CODEBLOCK1

Respect the state machine. Never skip from IDLE to RUNNING.