Graceful Error Recovery

Design review/orchestration flows so failures, missing dependencies, and partial automation never corrupt state or silently drop work.

Apply these standards: 1) Validate any persisted or resumed state/config before acting.

• If mode/settings are invalid, fall back to a safe fresh scan instead of trusting state. 2) Define a structured output contract and validate it before downstream use.
• Persist findings to an artifact (e.g., review-findings.json) and require downstream steps to validate required fields/types. 3) Never silently discard “dismissed” or “undecidable” outcomes in auto paths.
• If an automated decision can’t be made unambiguously, set the finding aside with an explicit reason and ensure it is recorded in the ledger/artifact so later cycles don’t re-litigate it. 4) Implement explicit graceful degradation when subsystems are unavailable.
• Don’t stall: choose an inline/sequential fallback appropriate to {auto_mode} vs interactive mode. 5) Make side effects and exits precise.
• Only true terminal branches should trigger “exit/on_complete” behavior; non-terminal “display” paths must return control without firing hooks.

Example (state validation + schema contract):

// Validate resume mode
const allowedModes = new Set(['interactive','auto','validate']);
if (!allowedModes.has(resumed.mode)) {
  // Safe recovery
  runFreshScan();
  return;
}

// Validate structured findings before use
const required = ['id','severity','type','summary','detail','file_line','proof'];
function validateFindingSchema(f){
  for (const k of required) if (!(k in f)) throw new Error(`Missing ${k}`);
  return true;
}

const findings = JSON.parse(fs.readFileSync('review-findings.json','utf8'));
findings.forEach(validateFindingSchema);
// proceed knowing the contract holds

Net effect: fewer corrupted/looping pipelines, no lost decisions, safer automation, and reliable recovery in the face of missing components or invalid persisted state.