RonanCodes/skills/in-your-face

In Your Face

Side questions get answered inline and then the session moves on. When the context clears, the answer is gone. This skill captures that answer as a standalone markdown file, renders it to HTML (plus PDF when it is free), and launches it on screen automatically. The launch is the point: an artifact nobody sees is as dead as the chat scroll it came from.

When to fire

• User says "in your face", "don't let me lose this", "save that answer", "make that durable", or asks for "that thing you told me earlier" to be saved.
• You just answered a substantive side question the user will plausibly want later: a recipe, a tool comparison, a how-to, a generated prompt for another AI, a decision walkthrough. Offer or fire proactively; the trigger is the answer's future value, not its length.

Do NOT fire for throwaway answers (one-liners, yes/no, ephemeral state like "what branch am I on") or for content that already lives in a durable home (a committed doc, an open PR, a wiki page).

Usage

/ro:in-your-face                       # capture the most recent side answer
/ro:in-your-face the docker networking explanation   # capture a specific earlier answer
/ro:in-your-face --vault ai-research   # force wiki routing to a vault
/ro:in-your-face --no-open             # write everything but skip the launch

Step 1: Capture

Write the answer as clean standalone markdown. Standalone means a reader with zero chat context understands it:

• Restate the original question as the opening (a title plus one or two framing sentences).
• Include code blocks, commands, and generated prompts verbatim, not summarised.
• Date it (YYYY-MM-DD near the top) so a stale copy is self-evidently stale.
• Apply /ro:write-copy rules: no em-dashes or en-dashes, no AI-tell vocabulary, no rhetorical-reversal filler.

Step 2: Route

In order:

1. Wiki vault, when the project maps to one. If ./.ronan-skills-llm-wiki.local.json exists (or /ro:wiki resolves a vault for this project), ingest the answer there following /ro:wiki conventions, and still write the local copy below. The vault is the durable memory; the local copy is what gets launched.
2. Local file otherwise. Home: ~/Desktop/in-your-face/<YYYY-MM-DD>-<slug>.md. Desktop on purpose: a human sees it without being told where to look. Create the directory if missing (mkdir -p).

Step 3: Render and launch

bash <skill-dir>/scripts/render-launch.sh ~/Desktop/in-your-face/2026-07-04-example.md

The script:

• Renders HTML beside the markdown file. Uses pandoc when installed; otherwise a built-in python fallback wraps the escaped text in a styled <pre> page. The launch never fails for lack of tooling.
• Produces a PDF only when it is trivially available (pandoc plus weasyprint or wkhtmltopdf already on PATH). It NEVER installs anything, never touches LaTeX, and never blocks the launch on PDF failure.
• opens the HTML automatically. That auto-launch is the skill's reason to exist; only suppress it with --no-open (tests, SSH sessions).

Step 4: Report

Print the three paths the script echoes (md / html / pdf-or-skipped) and, when wiki-routed, the vault page path too:

md:   /Users/ronan/Desktop/in-your-face/2026-07-04-docker-networking.md
html: /Users/ronan/Desktop/in-your-face/2026-07-04-docker-networking.html
pdf:  skipped (no pdf engine)
wiki: ~/Dev/ai-projects/llm-wiki/vaults/llm-wiki-ai-research/wiki/sources/docker-networking.md

Worked example

2026-07-04: a session produced a ChatGPT deep-research prompt for nutmeg model selection, plus the session answers that shaped it. It was hand-saved to ~/Desktop/nutmeg-model-deep-research-prompt.md, rendered with pandoc to the sibling .html, and launched with open. This skill mechanises exactly that flow: same Desktop landing, same pandoc render, same auto-launch, minus the hand-rolling.

Related

• /ro:wiki handles the vault ingest when the project is wiki-mapped; this skill defers to its conventions rather than re-implementing them.
• /ro:handoff overlaps on "session answers worth keeping", but serves resumable state. These artifacts are terminal (read them, done), so the pickup pattern (registered dir + printed pickup command) deliberately does NOT apply here.