Neeeophytee/finding-unknowns-skills

Finding‑Unknowns Skills – what it is

A small collection of 11 “skills” written in the SKILL.md format that many AI‑coding agents understand (Claude Code, OpenAI Codex, Kimi Code CLI, Hermes, Cursor, etc.). Each skill is a reusable command that helps a developer surface unknowns before they become costly implementation problems. The ideas come from Thariq Shihipar’s essays on “finding your unknowns” and on context‑engineering for Claude 5.

Category	Skills (short description)
Before you code	blindspot-pass – surface unknown‑unknowns; brainstorm-prototypes – generate throw‑away variations; interview-me – ask architecture‑changing questions; reference-hunt – locate existing code that can serve as a spec
During implementation	implementation-plan – prioritized plan; implementation‑notes – log deviations for future learning
After you finish	pitch‑packager – bundle spec, prototype and notes for reviewers; change‑quiz – generate a comprehension quiz before merging
Context‑level (affects every task)	context‑audit – clean up contradictions/duplicates in your agent’s system prompt; agent‑interface‑design – define safe tool interfaces; progressive‑disclosure – split large skills/specs so they load only when needed

The repo supplies the skill definitions, a one‑file “passive guidance” version (guidance/finding-unknowns.md), and detailed install scripts for each supported agent.

---

How to get them

# Recommended single command – works for Claude Code, Cursor, Codex, Gemini, Kimi, etc.
npx skills add Neeeophytee/finding-unknowns-skills

Add --list to see the 11 skills first, or --skill blindspot-pass to install just one.

Agent‑specific options

• Claude Code – also available as a marketplace plugin:

  /plugin marketplace add Neeeophytee/finding-unknowns-skills
  /plugin install finding-unknowns@finding-unknowns-skills
  

• OpenAI Codex – copy the skills/ folder into ~/.agents/skills/ or install via the Codex plugin commands shown in the README.
• Hermes – point ~/.hermes/config.yaml at the repo’s skills/ directory.
• Kimi Code CLI – autodiscovered if placed in any of the standard skill directories.
• Manual – drop the desired skills/<name>/ folder into ~/.claude/skills/, ~/.codex/skills/, or the project‑local .claude/skills/.

If you prefer a non‑executable reference, copy guidance/finding-unknowns.md to the root of your repo as CLAUDE.md (Claude Code) or AGENTS.md (Codex and other agents).

---

When to use which skill

Situation	Recommended skill
You’re entering a new codebase or domain	blindspot-pass
You need quick visual prototypes to articulate a vague idea	brainstorm-prototypes
You have a rough concept but still have open questions	interview-me
You know the behavior you need exists somewhere else	reference-hunt
Ready to start building	implementation-plan (follow with implementation-notes)
You’ve built something and need stakeholder buy‑in	pitch-packager
Before merging, ensure the change is understood	change-quiz
Your agent’s system prompt has grown noisy or contradictory	context-audit
You’re exposing tools to the agent and want a safe interface	agent‑interface‑design
A single skill or spec is too large for the model’s context window	progressive-disclosure

---

What you get (no hidden magic)

• Plain‑text SKILL.md files – each contains a short description, trigger phrase, and optional flags (e.g., disable-model-invocation: true). Agents that understand the format automatically make the skill available in their UI or via slash‑commands.
• Cross‑agent compatibility – the same skill works in Claude Code, Codex, Kimi, Hermes, Cursor, etc., without rewriting code.
• MIT‑licensed text – you can copy or adapt the skill definitions freely; the underlying prompting techniques are attributed to the original essays.

---

Who might find this useful?

• Developers who rely on LLM‑powered coding assistants and want a systematic way to surface hidden assumptions before they cost time.
• Team leads looking for lightweight documentation/verification steps (pitch-packager, change-quiz).
• Prompt engineers who maintain a shared CLAUDE.md/AGENTS.md for a project and need tools to keep that context clean (context-audit).

---

Quick start example (Claude Code)

1. Install all skills:

   npx skills add Neeeophytee/finding-unknowns-skills
   

2. In Claude Code, type /blindspot-pass and paste a brief description of the new module you’re about to work on.
3. Claude will ask you probing questions, surface “unknown unknowns”, and suggest a refined prompt you can use for the next coding step.

---

Bottom line: This repo is a ready‑to‑use toolbox that turns proven prompting techniques into reusable, installable commands for a wide range of LLM coding agents, helping you catch hidden gaps early and keep your AI‑augmented workflow tidy.