Como usar
- Copie o comando e verifique o pacote antes de instalar no seu ambiente de agente.
- Consulte no original as ferramentas, credenciais e ambientes necessários.
- Descreva os dados de entrada e a entrega esperada; confira a saída com as fontes.
npx skills add https://github.com/jpeetz/agent-skills --skill finding-unknownsAntes de começar
- A instalação não inclui assinatura de modelo, franquia API ou acesso aos seus arquivos.
- Estrelas e instalações indicam atividade, não um teste independente de qualidade ou segurança.
Renderizado do repositório de origem, preservando títulos, exemplos, código, tabelas, links e imagens.
Finding Unknowns — Meta-Cognitive Skill Suite
A collection of 11 meta-cognitive skills that make coding agents surface unknown unknowns before they get expensive. These skills teach agents to discover what they don't know, audit their own context, design better interfaces, and structure implementation for maximum reviewability.
Based on Thariq Shihipar's "finding your unknowns" techniques. Originally published by Neeeophytee at https://github.com/Neeeophytee/finding-unknowns-skills.
Skills in This Suite
| Skill | Description |
|---|---|
| blindspot-pass | Surface unknown unknowns before work starts — landmines, hidden context, quality calibration |
| interview-me | One-question-at-a-time targeted interview to resolve remaining ambiguity |
| reference-hunt | Extract semantics from existing code as specification, then reimplement |
| implementation-plan | Write plans ordered by likelihood of tweaking, not build order |
| implementation-notes | Log deviations from plan and discovered edge cases during a build |
| pitch-packager | Package work into a single review-ready document that gets fast approvals |
| change-quiz | Produce a diff report + 5-8 question quiz the user must pass before merge |
| brainstorm-prototypes | Generate wildly different throwaway variations for user reaction |
| context-audit | Audit CLAUDE.md, AGENTS.md, skills for contradictions, bloat, and overlap |
| agent-interface-design | Design MCP servers, tool definitions, and CLIs that teach their own use |
| progressive-disclosure | Split oversized skills/specs into entry + on-demand files |
When to Use
- You're about to work in an unfamiliar codebase area or domain
- You're planning a complex multi-step implementation
- You've just finished a long session and need to verify what changed
- User asks to "audit context" or "find my blindspots"
- You're designing an MCP server or agent-facing tool
Hermes Installation
This suite can be installed as external skills. Add to your ~/.hermes/config.yaml:
skills:
external_dirs:
- /path/to/finding-unknowns/skillsThen run hermes skills list to verify all 11 skills are enabled.
See INSTALL-HERMES.md in the original repository at https://github.com/Neeeophytee/finding-unknowns-skills for alternative installation methods.
Skill Index
Each sub-skill is in skills/<name>/SKILL.md. Refer to the individual skill file for full procedure.
- blindspot-pass — Systematic unknown-unknowns discovery
- interview-me — One-question-at-a-time ambiguity resolution
- reference-hunt — Code-as-specification reimplementation
- implementation-plan — Decision-first plan authoring
- implementation-notes — Running deviation log
- pitch-packager — Review-ready work packaging
- change-quiz — Pre-merge comprehension gate
- brainstorm-prototypes — Throwaway variation generation
- context-audit — Agent context rightsizing
- agent-interface-design — Self-teaching tool interfaces
- progressive-disclosure — On-demand content structuring
Pitfalls
progressive-disclosureusesdisable-model-invocation: truein original source (invoked only by user, not auto-activated). In Hermes, it behaves as model-invoked. Keep this in mind for context cost accounting.- The suite is most effective when used as a cycle: Context Audit → Brainstorm → Interview → Plan → Implement (with Notes) → Quiz → Pitch. Not all steps are needed for every task.
- Individual skills work standalone — you can pick just the ones you need.
Verification
- Run
hermes skills listand verify all 11 skill names appear as "enabled". - Give the agent a task in an unfamiliar domain and check if
blindspot-passtriggers. - After a long session, ask "what did we change?" and verify
change-quizproduces a comprehensive report. - Request a context audit on your active configuration and verify contradictions are surfaced.
