insights-pipeline

@cianos95-dev/insights-pipeline

0 forks

Updated 4/1/2026

Guide for archiving Claude Code Insights HTML reports as structured Markdown and extracting actionable patterns to improve CLAUDE.md and workflows. Use when archiving an Insights report, reviewing past archives, extracting CLAUDE.md improvement candidates, or comparing trends across reports. Trigger with phrases like "archive insights report", "review insights", "insights trend", "what did insights suggest", "insights to CLAUDE.md".

Installation

$npx agent-skills-cli install @cianos95-dev/insights-pipeline

Claude Code

Cursor

Copilot

Codex

Antigravity

Details

Repositorycianos95-dev/spec-driven-development

Pathskills/insights-pipeline/SKILL.md

Branchmain

Scoped Name@cianos95-dev/insights-pipeline

Usage

After installing, this skill will be available to your AI coding assistant.

Verify installation:

npx agent-skills-cli list

Skill Instructions

name: insights-pipeline description: | Guide for archiving Claude Code Insights HTML reports as structured Markdown and extracting actionable patterns to improve CLAUDE.md and workflows. Use when archiving an Insights report, reviewing past archives, extracting CLAUDE.md improvement candidates, or comparing trends across reports. Trigger with phrases like "archive insights report", "review insights", "insights trend", "what did insights suggest", "insights to CLAUDE.md". compatibility: surfaces: [code, cowork] tier: degraded-cowork degradation_notes: "HTML report archival requires local filesystem; Cowork can access archived reports via Linear"

Insights Pipeline

Archive Claude Code Insights reports as structured Markdown and extract patterns to improve your CLAUDE.md and workflows.

This is a methodology skill — it guides the agent through archiving and analysis using standard file tools (Read, Write, Edit, Grep, Glob). There is no runtime or database involved.

What This Skill Does

Archives HTML Insights reports to ~/.claude/insights/YYYY-MM-DD.md as structured Markdown.
Extracts actionable patterns: friction points, CLAUDE.md suggestions, feature recommendations, workflow patterns.
Compares current report against prior archived reports to identify improvement trends.

Archive Format

Each archived report follows this structure:

---
source: Claude Code Insights (Anthropic)
period: YYYY-MM-DD to YYYY-MM-DD
messages: N
sessions: N
archived: YYYY-MM-DD
original: [path to source file]
version: 1
---

# Claude Code Insights — [Period]

## At a Glance
## Project Areas
## Usage Stats
## Big Wins
## Friction Points
## CLAUDE.md Suggestions
## Features Recommended
## Patterns to Adopt
## On the Horizon
## Outcomes
## Satisfaction

Storage

Location: ~/.claude/insights/
Naming: YYYY-MM-DD.md where the date is the end date of the report period.
Idempotent: Re-running on the same report produces the same output. Do not overwrite an existing archive unless the user explicitly requests it.

Extraction Rules

When converting HTML reports to Markdown:

Strip all CSS and JavaScript. Extract text content only.
Preserve data fidelity. Every number, percentage, and metric from the original appears in the archive.
Use tables for structured data. Charts become tables. Bar charts become | Label | Value | tables.
Keep code suggestions verbatim. CLAUDE.md additions and prompt templates are preserved exactly.
Summarize narratives. Multi-paragraph prose sections become 2-3 sentence summaries.

Pattern Extraction

After archiving, extract these actionable outputs:

CLAUDE.md Candidates

Any suggestion from the report's "CLAUDE.md Suggestions" section.
Any friction pattern that occurred 3+ times (indicates a missing rule).
Any environment assumption error (indicates missing context in CLAUDE.md).

Skill Candidates

Any repeated multi-step workflow mentioned in "Features Recommended" or "Patterns to Adopt".
Any workflow the user performs manually that could be codified as a skill.

Trend Comparison

When multiple archived reports exist in ~/.claude/insights/, compare them using patterns.json (produced by the pattern-aggregation skill):

Are wrong-approach friction counts decreasing? (Check trend field in patterns.json)
Is the ratio of fully-achieved outcomes improving?
Which CLAUDE.md suggestions were adopted and did friction decrease afterward?

For structured cross-session trend analysis, run aggregate-patterns.sh (from the pattern-aggregation skill) to rebuild ~/.claude/insights/patterns.json from all archived reports. This provides normalized friction types, per-report counts, and trend calculations. For quick single-report comparisons, reading the archived Markdown files directly is sufficient.

Prerequisites

Claude Code Insights report (HTML format from Anthropic).
Write access to ~/.claude/insights/.

/ccc:insights command — runs the archive-and-learn cycle with --archive, --review, --trend, and --suggest modes.
CLAUDE.md — destination for extracted rules.
pattern-aggregation skill — cross-session friction pattern aggregation, produces patterns.json from archived reports.

More by cianos95-dev

View all

prfaq-methodology

Working Backwards PR/FAQ methodology for spec drafting with 4 templates, interactive drafting guidance, and structured questioning techniques. Use when writing a new spec, choosing a PR/FAQ template, drafting a press release, defining acceptance criteria, or structuring a feature proposal with pre-mortem analysis. Trigger with phrases like "write a spec", "PR/FAQ for this feature", "working backwards document", "which template should I use", "draft a press release", "pre-mortem analysis".

project-cleanup

One-time project normalization: reclassify issues vs documents, enforce naming conventions, migrate deprecated labels, delete non-actionable items, and create structural project documents. Use when onboarding a new project to CCC conventions, after importing issues from another tracker, or when a project has accumulated convention debt (wrong issue types, bracket prefixes, missing type labels). Trigger with phrases like "clean up this project", "normalize project issues", "apply CCC conventions", "convert research notes to documents", "fix bracket prefixes", "project cleanup".

context-window-management

Context window management strategies for multi-tool AI agents. Covers a 3-tier delegation model for controlling what enters the main conversation, context budget thresholds, subagent return discipline, and model mixing recommendations. Prevents context exhaustion during complex sessions. Use when planning subagent delegation, managing long sessions, deciding what to delegate vs handle directly, or choosing model tiers for subtasks. Trigger with phrases like "context is getting long", "should I delegate this", "subagent return format", "model mixing strategy", "context budget", "session splitting", "when to use haiku vs opus".

tdd-enforcement

Enforce TDD red-green-refactor discipline during CCC Stage 5-6 implementation. Derives test cases from spec acceptance criteria and PR/FAQ documents rather than generic test suggestions. Blocks implementation code before a failing test exists. Tracks cycle state across the session and integrates with .ccc-state.json task context. Use when implementing features with testable acceptance criteria in the CCC workflow, when the execution mode is exec:tdd, or when you need to enforce test-first discipline. Trigger with phrases like "enforce TDD", "red green refactor", "test first", "write a failing test", "no implementation yet", "TDD cycle", "exec:tdd mode", "derive tests from spec", "acceptance criteria to tests".