Codex Insights Skill

Generate high-signal Codex usage insights from local Codex data (state_5.sqlite + rollout files), then compile a report with visualizations.

This repository contains a complete skill bundle at .agents/skills/codex-insights with:

• extraction (manifest, meta, stats)
• rule-based facet analysis (default, optional hybrid rules + LLM)
• narrative synthesis (LLM, strict JSON contract)
• report rendering (report.html)

Install with npx -y skills add JacobLinCool/codex-insights, then run it using $codex-insights in Codex CLI or Codex App.

What This Skill Produces

Given Codex local state, the pipeline generates:

• manifest.jsonl + manifest.meta.json
• meta/<session_id>.json (SessionMetaV1)
• evidence/<session_id>.json (privacy-filtered evidence)
• facets/<session_id>.json (FacetV1)
• stats.json (StatsV1)
• narrative.json (NarrativeV1)
• report.html
• state_index.json (incremental rebuild state)

Data Inputs

Required inputs:

• SQLite DB: ~/.codex/state_5.sqlite
• Sessions root: ~/.codex/sessions
• Archived sessions root: ~/.codex/archived_sessions

Default output workspace:

• ~/.codex/insights/latest

Requirements

• Python 3.10+
• codex CLI available in PATH for hybrid classification and narrative stages
• Access to local Codex data paths listed above

Quick Start

Run the full pipeline:

python3 .agents/skills/codex-insights/scripts/run_pipeline.py \
  --db ~/.codex/state_5.sqlite \
  --sessions-root ~/.codex/sessions \
  --archived-root ~/.codex/archived_sessions \
  --scope sqlite_threads \
  --privacy redacted \
  --engine rules_only \
  --classifier-model gpt-5.3-codex-spark \
  --narrative-model gpt-5.3-codex

Then open:

~/.codex/insights/latest/report.html

Screenshots

Overview + KPI Cards

Activity, Model Trends, and Outcome Composition

Concurrency and Operational Diagnostics

Narrative Sections