cyberneticlibrary

Execute epic with budget tracking

yolo-epicworkflowsetup L30
Sheldon-92/TAD
What it does

Execute Epic phases with design review, implementation, and verification

Best for

Executing a multi-phase Epic with parallel design/implementation review and human checkpoints at budget/escalation gates.

Inputs
  • · epic_path: path to Epic definition file
  • · epic_slug: short name for evidence storage
  • · phase_number: integer (1+)
  • · phase_name: human readable phase name
  • · handoff_path: where to write design handoff
  • · completion_path: where to write implementation completion
  • · grounding_path: optional override for grounding file
  • · reviewer_count: number of parallel reviewers (default 2)
  • · steps: array of 'design', 'review', 'implement', 'impl_review' to run
Outputs
  • · HANDOFF.md with YAML frontmatter (task_type, e2e_required, research_required, git_tracked_dirs, Acceptance Criteria)
  • · COMPLETION.md with files_changed, commit_message, layer1_passed verdict
  • · evidence files (design review, impl review findings with P0/P1/P2 counts)
  • · budget consumption report
Requires
  • · Epic template files (.tad/templates/handoff-a-to-b.md, completion template)
  • · Conductor task system
  • · parallel reviewers (Alex, Blake agents)
Preconditions
  • · Epic file exists at epic_path with Phase N detail block
  • · Grounding file exists (or created from git state)
  • · Handoff template available at .tad/templates/handoff-a-to-b.md
  • · Steps array is non-empty and contains valid step names
Failure modes
  • · First design attempt <50 lines → re-spawn Alex design agent (loop until >= 50 lines)
  • · Reviewer P0 findings block implementation → Blake waits for resolution
  • · Implementation exceeds budget → escalation to human checkpoint before continuing
  • · layer1_passed=false → implementation reviewers notified, evidence added to task
Trust signals
  • · YAML frontmatter enforces structured handoff (task_type, requirements, acceptance criteria)
  • · Design must be >= 50 lines (enforces completeness before review)
  • · Parallel reviewers (not sequential) reduce phase latency
  • · Budget reporting at each checkpoint (tokens, escalations visible to human)
  • · Evidence stored per epic_slug for review audit trail