Content
61%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-organized, appropriately lean guide whose main weaknesses are incomplete executable guidance (pseudocode detection/JIT sections, unconstructed attacker transactions) and a missing simulation/validation checkpoint before bundle submission. Structure and conciseness are solid.
Suggestions
Add executable victim-detection code (real mempool filtering for swapExactTokensForTokens/exactInputSingle and the slippage-vs-gas profitability check) instead of comment-only pseudocode.
Insert an explicit numbered workflow with a validation step: simulate the bundle (e.g., fb.simulate()) and verify profitability before sendRawBundle, with a fix-and-retry loop on simulation failure.
Show how attackerTx1/attackerTx2 amounts are computed (front-run size, expected back-run proceeds) or point to a reference file with the full working bot.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Sections are compact with lean code blocks and no padding explaining concepts Claude already knows; the one-line sandwich definition is justified. Not 5 because a few sentences (e.g., in 'OPSEC / ethics / legal' and 'Common pitfalls') restate what code comments and earlier sections already convey. | 4 / 5 |
Actionability | The Flashbots JS bundle-submission snippet is near-executable, but the detection section is comment-only pseudocode ('Filter for Uniswap V2 swapExactTokensForTokens...'), the JIT section is pure comment pseudocode, and the core attacker transactions (attackerTx1/attackerTx2, including the swap-amount math) are never constructed. This matches the anchor 'pseudocode instead of executable code; missing key details' rather than the mostly-executable anchor 4. | 3 / 5 |
Workflow Clarity | An implied sequence (detect → construct → submit) is legible via section ordering, but steps are not explicitly numbered and there are no validation checkpoints — critically, no simulate-before-send step (e.g., Flashbots simulate()) for an operation that spends real gas and capital, which the rubric caps at 3. Not 2 because the sequence itself is clear and the pitfalls/defenses sections add process guidance. | 3 / 5 |
Progressive Disclosure | No bundle files exist and the ~105-line body is organized into clear, well-scoped sections (detect, construct, JIT, multi-block, defenses, pitfalls, tooling, OPSEC, references) with content appropriately placed for a single-file skill. Not 5 because it exceeds the under-50-line simple-skill exception and topics like Flashbots API usage or tooling setup would naturally belong in one-level-deep reference files. | 4 / 5 |
Total | 14 / 20 Passed |