New Skills! v1.2 brings /wait-what, /writing-for-agents, and fixes /grill-me

Tech/Product updates (Skills repo v1.2.0)

New documentation site

• Adds “proper documentation” for the Skills project.
• Site: aihero.dev/skills
• Navigation/content structure follows a main flow like: grill with docs → spec → tickets → implement → code review
• Includes:
  • A full reference panel for every skill
  • A common questions section, sourced from the creator’s maintained personal wiki
• Skills are linked to an AI coding dictionary (e.g., the “ticket” skill references a dictionary definition).

Skills now distributed via Claude Code plugin marketplace

• The project becomes an official part of the Claude Code official plug-in marketplace.
• Users can install the plugin directly in Claude Code via search (e.g., by creator name).
• Installation is presented as a read-only bundle, with updates pulled automatically when the creator updates the skills.

Codeex compatibility via OpenAI-style YAML sidecars

• For Codeex users, the release adds OpenAI.YAML files for every skill, so skills should work “out of the box” with UIs like Codeex UI.
• Specifically fixes/handles implicit invocation behavior by shipping sidecar files to support allow_implicit_invocation: false with Codeex 2.
• Context:
  • Previously, there were “model invoked” vs “user invoked” skills.
  • User-invoked skills were hidden from the agent context window in some harnesses.
  • They did not work for Codeex until these sidecar files were added.

---

New/updated skills

New skill: wait what?

• Created to address a verbosity/readability issue with Opus / Opus 5 (overly verbose, hard to read, odd phrasing).
• The skill instructs the agent to:
  1. Use Simple Technical English (ASD-STE-100) to force clearer, simpler declarative sentences.
  2. Ground responses in context.md using the project’s “ubiquitous language.”
• Intended usage: run it when the agent outputs “random garbage” and you don’t understand what it said, and it should produce a better alternative.

Update to grill me

• Previously: one question per turn (agent asks one, user answers, repeat).
• Problem: near the end of a grilling session, remaining questions can be easy/repetitive, causing a slow, dead-feeling experience.
• New behavior: multiple questions per turn, while avoiding dependency ordering issues.
• Implementation concept:
  • Questions form a graph
  • The skill answers only questions “available now,” using rounds:
    • Round 1: earliest/critical questions
    • Round N: subsequent questions unlocked by earlier answers
• UI detail: shows rounds and recommended answers with an eye-friendly presentation (including a mention of emojis for navigation).

Update/rename concept: writing for agents

• Replaces/expands the previous “writing great skills” usage.
• The creator noticed they were using the old skill for broader purposes beyond skills documentation:
  • improving how agents read configs
  • making output more concise/predictable
• The new skill supports applying writing rules to files like agents.mmd and claude.md (as mentioned), and during creation/editing of skills.
• Also helps extract/pull out writing style into skills to avoid “horrible frontloading.”
• Marked as model invocable, so it can be automatically used when modifying agents.mmd.

New skill: wizard

• Generates an interactive bash “wizard” that walks humans through setup steps they can perform.
• Example use case: provisioning/migrating to a remote box setup (AWS context mentioned).
• Key properties:
  • Deterministic script flow (explicitly not using the agent to execute cloud actions via Anthropic/LLM “computer use”)
  • Handles navigation steps, login, prompting for specific changes, copying/pasting API keys
  • Saves required artifacts into the filesystem and can write values into GitHub secrets
  • Uses a staged UI until provisioning completes
• Claimed result: turns painful provisioning into a “joyful” process.

New skill: two questionnaire

• Purpose: extract the decisions made during a grilling session into a Markdown document for collaboration.
• Example: building a garden office where the creator’s wife needed to participate in the design.
• Workflow:
  1. Agent grills in a session
  2. Agent outputs a structured markdown “questionnaire” document
  3. Shared via Google Docs for discussion/comments
  4. Answers are fed back into the agent
• Motivation: agents are hard to collaborate with directly across teams unless integrated into chat workflows (Slack/Teams). This enables portable, documentation-based collaboration.

---

Other notes / release info

• Notes additional small changes in GitHub Releases under the v1.2.0 changelog.
• Promotes an AI coding crash course (self-paced, cheaper than previous cohort-style paid courses) to provide a stable foundation despite expected ongoing model changes.

---

Main speaker / source(s)

• Primary speaker: The creator/maintainer of the “Skills” repo (also referenced via plugin marketplace search as “Matt Pocco / Matt PCO”).
• Primary content source: The Skills repository documentation and release v1.2.0 (at aihero.dev/skills and in GitHub releases).