LLMSkillHub
mirroredCopied from PlamenTSV/plamen under MIT. PlamenTSV has not claimed this page, and is not affiliated with LLM SkillHub.

plamentsv / event-correctness

Trigger 15 events detected in recon event_definitions.md (optional skill) - Used By breadth agents (assigned to core state or dedicated agent)

0
SKILL.md
2.4 KB

Skill: EVENT_CORRECTNESS

Trigger: >15 events detected in recon event_definitions.md (optional skill)
Used By: breadth agents (assigned to core state or dedicated agent)
Purpose: Verify emitted event parameters match actual state changes

Methodology

Step 1: Inventory All Emit Statements

From {SCRATCHPAD}/emit_list.md, extract every emit statement with:

  • Event name, parameter values passed, source location
  • The state-changing operation(s) that precede the emit

Step 2: Parameter Semantic Check

For EACH emit statement, verify:

#CheckQuestion
1Value accuracyDoes each parameter reflect the ACTUAL post-operation state? (not a stale pre-operation value, not an input parameter that was modified before use)
2Index correctnessIf the event indexes an entity (ID, address, index), is the index the CORRECT entity? (not off-by-one, not a different entity's ID, not a loop variable after increment)
3OrderingIs the emit placed AFTER all state changes it describes? (not before a conditional that could change the values)
4Conditional coverageIf the function has branching logic, does EVERY branch that modifies state emit the appropriate event? (no silent state changes)
5Parameter countDo the emitted parameters match the event definition? (Solidity allows emitting fewer params - missing params default to zero)
6Semantic correctnessDoes each emitted variable match the SEMANTIC INTENT of the parameter name? If the event parameter is named tokensReceived, is the emitted value the actual tokens received (output), or is it the input amount (e.g., DAI spent)? Compare the parameter name against the variable being emitted - a mismatch between name semantics and actual value is a finding even if types match.

Step 3: Off-Chain Impact Assessment

For events consumed by off-chain systems (indexers, frontends, monitoring):

  • If event parameters are wrong → off-chain state diverges from on-chain state
  • Severity: typically LOW-MEDIUM unless financial decisions depend on indexed events

Output

Findings use IDs [EVT-N]. Include the emit location, the incorrect parameter, and the correct value it should emit.