Skip to content

Releases: ericfitz/skills

Release list

boring v0.1.0

Choose a tag to compare

@ericfitz ericfitz released this 05 May 04:13

First release of the boring writing-evaluation skill.

A skill that evaluates technical business writing — executive briefs, RFCs, status updates, finding writeups, architecture docs, proposals, technical reports — for "boringness" across a 20-sub-dimension taxonomy on four axes: Direction, Density, Texture, Surprise. Grounded in the MAC model of boredom (Westgate & Wilson, 2018) plus the craft tradition (Gopen-Swan, Williams, Provost, Minto).

What's in this release

  • Phase 1 — mechanical analyzer: 15 sub-dimensions checked deterministically via a Python script (spaCy + proselint + textstat + pypdf + python-docx). Locator-rich JSON output.
  • Phase 2 — LLM-judged sub-dimensions: 5 sub-dimensions (D1.2 missing stakes, D1.5 flat tension, D2.5 obvious claims, D4.2 no vivid imagery, D4.3 no counterintuitive claims) handled by the host LLM via per-dimension rubrics shipped in the skill.
  • Phase 3 — merged output: combined JSON for downstream consumers; the host LLM produces a prioritized summary for the user.

Calibration thresholds are intuitive defaults; not yet tuned against a labeled corpus of business writing. See `boring/docs/calibration-findings-2026-05-04.md` in the repo for the first calibration attempt.

Install - Claude Code

  1. Download `boring-0.1.0.zip` from this release.
  2. Copy it to your skills directory and unzip:
    cp boring-0.1.0.zip ~/.claude/skills/
    cd ~/.claude/skills && unzip boring-0.1.0.zip && rm boring-0.1.0.zip
  3. Build the analyzer's Python venv (requires uv):
    cd boring && uv sync

Install - Codex

  1. Download `boring-0.1.0.zip` from this release.
  2. Copy it to your skills directory and unzip:
    cp boring-0.1.0.zip ~/.codex/skills/
    cd ~/.codex/skills && unzip boring-0.1.0.zip && rm boring-0.1.0.zip
  3. Build the analyzer's Python venv (requires uv):
    cd boring && uv sync

Next steps

Then invoke the skill from a Claude Code or Codex session — the agent reads SKILL.md and runs the analyzer for you.

Input formats: .md, .markdown, .txt, .docx, .pdf (text-extractable; image-only PDFs are rejected).