Phase Goldmask — Shared Reference

Phase Goldmask — Shared Reference

Common patterns for Goldmask investigation agents (Impact + Wisdom + Lore layers).

Goldmask Architecture

The Goldmask pipeline synthesizes findings across three layers:

Layer	Agents	Purpose
Impact	5 tracers (API, business-logic, data-layer, config-dependency, event-message)	WHAT changed and WHERE it ripples
Wisdom	wisdom-sage	WHY code was written (git archaeology, developer intent)
Lore	lore-analyst	Risk scores from quantitative git history analysis

The goldmask-coordinator merges all three layers into a unified GOLDMASK.md report.

Investigation Agent Lifecycle

1. TaskList() → find available tasks
2. Claim task: TaskUpdate({ taskId, owner, status: "in_progress" })
3. Read task context (diff spec, file list, plan references)
4. Execute analysis protocol (layer-specific)
5. Write findings to output_path
6. Mark complete: TaskUpdate({ taskId, status: "completed" })
7. Send Seal to team-lead

Lore Layer — Risk Scoring

The lore-analyst computes per-file risk scores using quantitative git metrics:

Metric	Weight	What It Measures
Churn rate	0.30	Change frequency (commits/month)
Defect correlation	0.25	Fix-commit density
Author concentration	0.20	Bus factor (ownership spread)
Recency	0.15	Time since last change
Complexity proxy	0.10	File size as complexity estimate

Risk Tiers

Tier	Score Range	Meaning
CRITICAL	80-100	Highest risk — frequent changes, many fixes, few owners
HIGH	60-79	Elevated risk
MEDIUM	40-59	Moderate risk
LOW	20-39	Lower risk
STALE	0-19	Rarely touched — may indicate abandonment

Output: risk-map.json with per-file scores, tiers, and co-change clusters.

Coordinator Synthesis Protocol

The goldmask-coordinator merges layer outputs using a 3D priority formula:

priority = 0.40 * impact + 0.35 * caution + 0.25 * risk

Synthesis Steps

1. Read Impact Layer outputs (5 tracer reports)
2. Read Wisdom Layer output (intent analysis)
3. Read Lore Layer output (risk-map.json)
4. Build correlation graph (file → [impact, caution, risk])
5. Compute 3D priority scores
6. Assess collateral damage (files affected by changes)
7. Detect swarm patterns (multiple findings converging on same area)
8. Double-check top 5 findings
9. Produce GOLDMASK.md + findings.json

Output Format

# GOLDMASK — Cross-Layer Impact Analysis

## Priority Matrix
| File | Impact | Caution | Risk | Priority | Category |

## Collateral Damage Assessment
{Files affected indirectly by changes}

## Swarm Detection
{Areas where multiple findings converge}

## Top Findings
{Prioritized by 3D formula}

Agent-Specific Content (NOT in this shared file)

• Individual tracer analysis protocols (API, business-logic, etc.)
• Git command sequences for lore-analyst metrics
• Guard checks G1-G5 (lore-analyst validation)
• Wisdom-sage intent classification taxonomy
• Co-change graph construction algorithm
• Percentile normalization details