Finding-Unknowns Skills
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
- 358 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skills are instructions and run nothing themselves
- context-audit reads your CLAUDE.md and skills to find contradictions in them
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/finding-unknowns-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills are instructions and run nothing themselves
- context-audit reads your CLAUDE.md and skills to find contradictions in them
This is third-party code. Review the repository files before installing.
What it does
A set of 14 skills for Claude Code, Codex and any agent that reads the SKILL.md format. A prompt describes a task only in part, and the gaps between it and the real code get expensive when they surface late. The skills make the agent find them up front: eight work on a single task, three clean up the context the agent carries into every task, and three more back conclusions with evidence. Before work starts, interview-me, blindspot-pass and brainstorm-prototypes clarify the task. reference-hunt turns working code in another language into a spec, and implementation-plan puts the decisions you are most likely to change at the top of the plan. During work, implementation-notes logs every deviation from the plan. Afterwards pitch-packager bundles the spec, prototype and notes for reviewers, and change-quiz checks that you understood the change before merging. context-audit, agent-interface-design and progressive-disclosure tidy the context. The first finds contradictions, duplicates and dead rules in CLAUDE.md and skills. Verification is covered by assumption-test, test-blindspots and regression-proof.
Who it is for. For developers who run large tasks through Claude Code or Codex and want to cut the cost of mistakes before the first line of code.
Good fit when
- A task is stated in broad terms and you need to settle the gaps one question at a time with interview-me
- You are entering an unfamiliar area and want to find what you do not know with blindspot-pass
- There is working code in another language and you want to use it as a spec through reference-hunt
- An agent fixed a bug, and regression-proof should show the regression test fails before the change and passes after it
- CLAUDE.md and skills have grown and picked up contradictions: run context-audit
Not a fit when
- You need a test framework or a test generator: the skills steer the agent's work and do not run tests themselves
- The task is small and clear at once: an interview and a plan only slow you down
- You want to change the agent's effort level: the skills do not affect it
Example request
Run interview-me with me on the payments migration task: architecture-changing questions first, then write the final decision list for the planLimitations
This is an independent community project; Anthropic did not publish it. Twelve skills are distilled from public essays by Thariq Shihipar of the Claude Code team, and the repository author added two more.
The disable-model-invocation field on progressive-disclosure makes the skill user-invoked only in Claude Code. Codex ignores the field: the author verified on v0.143 that the skill description still loads into the model-visible prompt.
In Cursor the install command discovers the skills, but the author has not yet confirmed invoking them there. The skills run nothing themselves; the agent runs tests and checks with your tools.
How to disable. If you installed through the plugin, disable it with /plugin in Claude Code. If you copied folders by hand, delete the relevant skills/<name>/ from your project's .claude/skills/ or from ~/.claude/skills/. The one-file variant is removed by deleting the block from CLAUDE.md or AGENTS.md.
FAQ
How do I install Finding-Unknowns Skills in Claude Code?
Two routes. npx skills add Neeeophytee/finding-unknowns-skills detects your agent and puts the skills in the right folder. Or use the plugin: /plugin marketplace add Neeeophytee/finding-unknowns-skills and /plugin install finding-unknowns@finding-unknowns-skills. By hand you can copy any folder from skills/ into .claude/skills/.
Do the skills work in Codex?
Yes, the SKILL.md format is the same and no conversion is needed. Install them with the same npx skills add command or copy the folders into ~/.agents/skills/. Codex also has a plugin: codex plugin marketplace add Neeeophytee/finding-unknowns-skills.
Which skill should I start with?
With interview-me: it asks one question at a time, architecture-changing ones first, and ends with a decision list for the plan. For an unfamiliar area use blindspot-pass, and for testing an assumption use assumption-test.
How is regression-proof different from test-blindspots?
regression-proof is for a known bug: it shows that the same regression test fails before the fix and passes after it. test-blindspots investigates what an already passing suite does not establish, using focused probes.
Is there a version without commands, in one file?
Yes, guidance/finding-unknowns.md. Copy it to your project root as CLAUDE.md for Claude Code or as AGENTS.md for Codex and other agents, or append it to your existing one. It works as passive guidance, with no commands to invoke.
Discussion
No questions or reviews yet.
Related
SkillsA set of Claude Code utilities: background skills, expert agents, slash commands and prompt templates for development
110 skills for Claude Code and Codex and seven Claude Code agents: development, product, UI, diagnostics and operations, installed whole or as a bundle
A set of Claude Code skills for the dev workflow: worktree isolation, a service layer, evidence and review
Claude Code Skills and Agents
Claude Code Skills & Agents
A Claude Code plugin bundling skills and sub-agents: deep research, image generation, GitHub work, code review and model switching
Eleven skills for Claude Code and Codex: test audit, git worktree cleanup, plan pages, screen recording transcripts and 1Password sign-in
Softaworks Agent Toolkit
Softaworks Agent Skills
A marketplace of skills, subagents and commands for coding agents: development, documentation, planning and workplace communication
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Finding-Unknowns Skills (this entry) | no | 358 | |
| Skills for real engineers by Matt Pocock | no | 306,265 | Matt Pocock's skill set for everyday engineering work |
| Spec Kit | no | 143,920 | GitHub's official toolkit for spec-driven development with commands and templates |
README in short
The description is built around a table of known and unknown and a list of skills by phase: before, during and after work. A separate section explains installation for Claude Code, Codex, Cursor and Hermes, plus the one-file variant. Links lead to examples, a compatibility table and a roadmap.
README badge
Original SKILL.md
--- name: interview-me description: Interview the user one question at a time to resolve remaining ambiguity before implementation. Use when planning or brainstorming is done but unknowns remain, or when the user asks to be interviewed about a task or spec. --- # Interview me Brainstorming is over and there are still gaps between the user's map and the territory. Close them by asking, one question at a time, starting with the questions whose answers would change the most. ## Steps 1. Read everything already established: the request, any spec, any prototypes, relevant code. Do not ask about things that are already answered. 2. Build a private list of open ambiguities and sort by blast radius: - **First: architecture-changers** — answers that would alter the data model, the interfaces, or the overall approach. - **Then: behavior definers** — edge cases, failure modes, defaults, permissions. - **Last: polish** — naming, copy, cosmetics. Often not worth asking; propose and move on. 3. Ask exactly one question per turn. For each: give the context that makes it matter, offer 2-3 concrete options with your recommendation, and accept "you decide" as an answer you then own. 4. Every few questions, checkpoint: restate what has been decided so far in one tight list, so drift dies early. 5. Stop when the remaining unknowns are cheaper to discover during implementation than to ask about now, and say that out loud. End with the final decision list, ready to paste into a plan. ## Guardrails - One question at a time means one. No question bundles. - Never ask a question whose answer is discoverable from the codebase; go look instead. - If an answer contradicts an earlier decision, flag the conflict immediately rather than silently taking the newest answer.