insightful

A collection of reusable CodeBuddy skills for evidence-based technical analysis and knowledge synthesis.

What's inside

codebase-insight

A deep codebase-cognition skill aimed at engineers, architects, and product owners. Instead of summarizing the directory tree, it rebuilds an evidence-based model of the target repository from its code, configuration, database, tests, and deployment artifacts:

• Product & business — what the system does and what problem it solves.
• Architecture — system boundaries, modules, dependencies, tech stack, and design patterns.
• Core flows — business chains, state transitions, and data lifecycles.
• Source deep dives — selected mechanisms explained through real control flow, state, failures, design tradeoffs, and reproducible experiments.
• Secondary development — extension-point maps, versioned API cards, minimal vertical examples, tests, and troubleshooting.
• Evidence conflicts — gaps between docs, comments, config, and actual implementation.
• Engineering review — highlights, hard parts, risks, lessons, and inspirations.
• Roadmap & refactoring — future plans and incremental refactor designs.
• Resume & interview — questions/answers grounded only in real code.

All conclusions are tagged by confidence (FACT / INFERENCE / UNKNOWN / PROPOSAL) and backed by file-path evidence, so the output can be verified rather than trusted.

The skill ships a runnable helper that produces a deterministic file inventory used as reconnaissance input:

python3 skills/codebase-insight/scripts/repo_inventory.py <repo-path> -o repo-inventory.json

It also includes a report validator for evidence-ledger structure, source-path and line references, confidence gates, and contribution attribution:

python3 skills/codebase-insight/scripts/validate_report.py <report-dir> --repo <repo-path>

video-knowledge-distiller

A local video knowledge-distillation skill for turning a creator's videos and subtitles into an installable consulting skill. It supports three intensities:

• Subtitle-first — read the full transcript and inspect frames only when the text is ambiguous or visually dependent.
• Visual cross-check — compare key chapters, code, diagrams, and demos with timestamped screenshots retained in the Markdown output.
• Research verification — additionally verify terminology, factual claims, version changes, disputes, and outdated advice against current primary sources.

Its output includes one Markdown document per video, cross-video domain syntheses, a collection overview, correction/update and evidence ledgers, timestamped screenshots, and a generated SKILL.md for later consultation. Deterministic helpers inventory media pairs, slice SRT/WebVTT by time range, and extract precisely named frames with FFmpeg.

Repository layout

skills/
├── codebase-insight/           # Source-code repository cognition
├── apk-insight/                # Android APK product/architecture analysis
└── video-knowledge-distiller/  # Video-to-consulting-skill distillation

Usage

Install the skill into your CodeBuddy skills directory and invoke it (codebase-insight) against any repository you want to understand. Modes:

• Quick — first contact: what it is, how to run it, key entry points.
• Full — architecture review, handover, refactor, due diligence.
• Topic — drill into one chain, service, module, or risk.

Each scope can combine architecture, source-dive, secondary-development, engineering-evolution, interview, and external-research tracks. Only requested or evidence-justified tracks are generated.

Invoke video-knowledge-distiller with a local media directory, an output directory, and optionally an intensity level (1–3). When omitted, the skill uses level 2 and records that choice in the generated knowledge base.

License

See individual skill directories for licensing terms.