Epic Blitz

Epic Blitz

A three-round pipeline for rigorous epic critique and revision. Fifteen independent critics across three abstraction levels (high, mid, low) review the epic in panels of five. After each panel, a senior reviser adjudicates the findings — accepting, rejecting, deferring, clarifying, or escalating — and produces a revised epic. The final readiness stage assesses whether the epic is ready for megaplan chain decomposition into milestone briefs.

When to invoke Epic Blitz

Use Epic Blitz when:

• You have a draft epic document and want adversarial review before committing to chain planning.
• You want to surface strategic risks, missing abstractions, decomposition flaws, convention mismatches, and implementation feasibility gaps.
• You want a structured, artifact-oriented critique process rather than ad-hoc human review.
• You need a readiness assessment before running megaplan chain.

Do NOT use Epic Blitz for:

• Prose or document review (use writing-panel-strict for that).
• Sprint-level critique (Epic Blitz is for epic-level documents).

Required input

• draft (file, required): Path to the epic markdown draft to review.

Usage

# Run with default profile
megaplan run epic-blitz path/to/epic.md

# Run with explicit inputs flag
megaplan run epic-blitz --inputs draft=path/to/epic.md

# Run with a specific profile
megaplan run epic-blitz path/to/epic.md --profile @epic-blitz:standard

Pipeline flow

Epic Blitz runs three critique-and-revision rounds, each at a different abstraction level:

Round 1: High abstraction

1. High Panel (parallel) — five critics review the draft:
   • existing_system_reuse — does the repo already solve this?
   • conceptual_fit — does this belong in megaplan's model?
   • missing_abstraction — is there a shared abstraction opportunity?
   • epic_decomposition — are milestones sliced correctly?
   • strategic_risk — is this solving the right problem?
2. High Revise — senior reviser adjudicates findings and produces a revised epic.

Round 2: Mid abstraction

3. Mid Panel (parallel) — five critics review the revised epic:
   • codebase_convention_fit — does the approach match existing patterns?
   • data_artifact_model — are files and schemas shaped correctly?
   • orchestration_semantics — do phase transitions and failures make sense?
   • agent_model_assignment — are the right models on the right jobs?
   • blast_radius — what could regress?
4. Mid Revise — senior reviser adjudicates findings and produces a further revised epic.

Round 3: Low abstraction

5. Low Panel (parallel) — five critics review the revised epic:
   • implementation_feasibility — can an agent execute without guessing?
   • testability — are concrete tests specified?
   • edge_cases — what about empty findings, malformed output, interrupted runs?
   • cli_ux_details — are names, flags, and errors clear?
   • migration_backcompat — does this preserve existing behavior?
6. Readiness — terminal reviser produces the final epic and chain-readiness assessment.

Expected artifacts

After a successful run, the plan directory contains:

<plan_dir>/
├── high_panel/
│   ├── existing_system_reuse/v1.md
│   ├── conceptual_fit/v1.md
│   ├── missing_abstraction/v1.md
│   ├── epic_decomposition/v1.md
│   └── strategic_risk/v1.md
├── high_revise/v1.md
├── mid_panel/
│   ├── codebase_convention_fit/v1.md
│   ├── data_artifact_model/v1.md
│   ├── orchestration_semantics/v1.md
│   ├── agent_model_assignment/v1.md
│   └── blast_radius/v1.md
├── mid_revise/v1.md
├── low_panel/
│   ├── implementation_feasibility/v1.md
│   ├── testability/v1.md
│   ├── edge_cases/v1.md
│   ├── cli_ux_details/v1.md
│   └── migration_backcompat/v1.md
├── readiness/v1.md
└── state.json

The terminal artifact is readiness/v1.md — the final revised epic with a chain-readiness assessment.

Profiles

Notes

• Epic Blitz v1 is fully non-interactive. All 15 critics and 3 revisers run without human gates.
• Each panel's critics receive the latest revised epic (not the original draft), ensuring later rounds build on prior revisions.
• Critics produce artifact-oriented findings with IDs, severity, rationale, evidence, and proposed actions.
• Revisers produce decision tables tracking every finding's disposition.