12 Factor Agents: How PromptKing Scores 11/12

The Framework

12 Factor Agents is an open-source framework for building reliable LLM-powered software. Core thesis: the most reliable agents in production are mostly deterministic code with small LLM steps at specific decision points. PromptKing is built on exactly this principle.

The One Gap: Factor 9

Factor 9 asks: when a tool call fails with an error (not loops — fails), does the system compact the error into the context window with a consecutive failure counter and escalate to a human at a threshold? PromptKing has cycle detection (loops). The named error event type with consecutive failure threshold is spec’d but not yet built.

The Planned Fix

error_type on trajectory_nodes (tool_failure / timeout / rate_limit / schema_violation / policy_rejection). consecutive_error_count per graph. governance_threshold configurable per org (default: 3). When threshold hit → same economic flag as cycle detection → Approval Required.

Why This Matters for M&A

An 11/12 score with a documented, scoped gap on the one remaining factor is a stronger diligence story than a vague claim of full compliance. The scorecard is the argument. The spec shows the path to closing it.

SCORECARD SUMMARY

SEE IT LIVE