Work Summarizer

Generates structured daily/weekly work reports from repo scans and transcript analysis, with optional Claude API synthesis.

Trigger

node dist/index.js { mode: "daily" | "weekly" }

Input Schema

interface SkillInput {
  mode?: "daily" | "weekly";              // Default: "daily"
  registryPath?: string;                  // Default: C:\dev\repo-registry.json
  outputDir?: string;                     // Default: C:\dev\CIP\CIC\logs\work-summaries
  includeRoutingArtifact?: boolean;       // Default: false
  reasoningEnabled?: boolean;             // Default: false
  anthropicModel?: string;                // Default: claude-haiku-4-5
  reasoningTimeoutMs?: number;            // Default: 30000
}

Output Schema

interface SkillOutput {
  status: "ok" | "error";
  message: string;
  data?: {
    schema_version: string;
    period: "daily" | "weekly";
    repos_scanned: number;
    modified_files: number;
    work_by_category: Record<string, number>;
    drift_signals: { count: number; score: number; summary: string };
    repo_deltas: Array<any>;
    subsystem_impacts: Array<any>;
    cross_repo_impacts: Array<any>;
    risks_or_followups: string[];
  };
}

Features

• Deterministic Pipeline: Git diff analysis + transcript scanning (no LLM required)
• 17-Category Taxonomy: Consistent file classification across all repos
• Weekly Aggregation: Loads last 7 daily reports; falls back to full scan if < 50% coverage
• Drift Detection: Keyword matching in transcripts (drift, driftScore, etc.)
• LLM Reasoning (optional): Per-category Claude API call with fallback on failure

Error Handling

Code	Message	Handler	Escalation
REGISTRY_NOT_FOUND	Registry JSON missing/empty	Fail	Create C:\dev\repo-registry.json
NO_MODIFIED_FILES	No changes detected	Fail	Run in test env with mock repos
OUTPUT_DIR_INVALID	Cannot write output	Fail	Check directory permissions

---

Full Reference

For Setup, Requirements, Configuration, Testing, Environment Variables:

→ See Skill Operator Guide

For examples, schema versioning, troubleshooting, and detailed workflows:

→ See docs/USAGE.md