Skills Collection

A bundle of skills

A curated, production-ready collection of Codex/Claude-style skills, packaged for easy hosting and reuse.

This repository is structured to be readable by both humans and agents: each skill lives in its own folder and is defined by a SKILL.md file with frontmatter metadata (name, description) and operational guidance.

Table of Contents

What this repository contains

This repo provides:

  • Reusable skill definitions for common agent workflows.
  • Consistent documentation style so users can discover and trust each skill.
  • Governance docs expected in professional open-source repositories (LICENSE, contribution guide, code of conduct, security policy).

If your goal is to host skills that feel polished like those on curated skill hubs, this structure gives you a strong baseline.

Boundary: this repository is the shared/global skill source. It must contain portable skills that are safe to propagate into multiple machines and client repos. Brain-owned orchestration skills, repo-specific Builder workflows, .builder/ routing, and client-specific bindings stay in their owning repo and fan out directly from that repo only after explicit approval. In particular, builder-pagecraft-html is a dans-brain project skill, not a global skills repo skill.

Repository structure

skills/
├─ README.md
├─ LICENSE
├─ CONTRIBUTING.md
├─ CODE_OF_CONDUCT.md
├─ SECURITY.md
├─ .gitignore
├─ sync-skills.sh    (propagate canonical skills → machine skill dirs)
├─ scripts/skillopt_pilot.py
│                   (no-mutate skill benchmark/proposal harness)
├─ tests/           (repo-level harness tests)
├─ agent-coordination/
│  └─ SKILL.md
├─ handoff/
│  └─ SKILL.md
├─ grill-me/
│  └─ SKILL.md
├─ parallel-dispatch/
│  └─ SKILL.md
├─ insight-lock/
│  └─ SKILL.md
├─ skill-drift-scanner/
├─ bogdan-baciu-design/
│  ├─ SKILL.md
│  ├─ references/brand-system.md
│  └─ assets/design-system/      (tokens, fonts, images, previews, UI kit)
├─ format-html/                  (multi-skill plugin bundle)
│  ├─ .claude-plugin/plugin.json
│  └─ skills/
│     ├─ format-html/            (overview + installer + shared design assets)
│     │  ├─ SKILL.md
│     │  ├─ install-pagecraft.sh
│     │  ├─ assets/css/pagecraft.css
│     │  ├─ assets/wrap-lab/     (good/bad wrap fixtures)
│     │  └─ references/          (tables, headers, text-wrap, number-formats, …)
│     ├─ verify-text-wrap/       (canonical home of the wrap probe + verifiers)
│     │  ├─ SKILL.md
│     │  ├─ runner.py · check-keystone.py · browserbase.py
│     │  └─ wrap-safe.css · wrapcheck.js   (wrap-safe runtime library)
│     ├─ table-system-migration/
│     │  ├─ SKILL.md
│     │  ├─ table-ratchet.py       (baseline ratchet for no-new table debt)
│     │  └─ table-ratchet-checklist.md
│     ├─ number-formats/
│     │  ├─ SKILL.md
│     │  ├─ formats.json         (byte-exact Excel format codes)
│     │  └─ apply-number-formats.py
│     └─ reskin/
│        ├─ SKILL.md
│        ├─ reskin.py            (detect → sync → reframe; hybrid bespoke/generic)
│        └─ reskin.example.json
└─ blog-image-gen/
   └─ SKILL.md

Bundles vs. loose skills. Most folders are single skills (one SKILL.md). format-html is a multi-skill plugin bundle: its members live under format-html/skills/ and stay self-contained (each owns its own tools, nothing vendored or duplicated). sync-skills.sh flattens bundle members on export, so each still installs as a top-level loose skill (~/.tool/skills/<name>/) invokable by its bare name.

Conventions

  • One skill per folder.
  • Required file in each folder: SKILL.md.
  • SKILL.md starts with YAML frontmatter:
    • name: stable slug.
    • description: explicit trigger/use-case text.
  • Keep skill instructions concrete, testable, and scoped.

Skill quality standard

Use this checklist before publishing a skill:

  • Trigger clarity: descriptions include concrete phrases users may type.
  • Scope boundaries: explicit non-goals (“what this skill does NOT do”).
  • Execution workflow: numbered steps with order dependencies.
  • Output contract: exact format expected from the agent.
  • Failure behavior: what to do when files, tools, or assumptions are missing.
  • Safety constraints: guardrails around destructive or high-risk actions.
  • Examples: at least one realistic input → output example.
  • Source attribution: include upstream references when adapted.

SkillOpt pilot

This repo includes a conservative SkillOpt-style pilot at scripts/skillopt_pilot.py. It does not call an LLM and does not mutate SKILL.md. It scores a skill document against deterministic benchmark rows, applies curated section patches to a copy of the body, and writes proposed.md plus receipt.json for human review.

The old excel-wow placeholder and its pilot benchmark were retired. New SkillOpt pilots should live under the skill they actually exercise, usually in <skill>/evals/skillopt_pilot.jsonl, and should be run against that skill's real SKILL.md.

Run the pilot tests with:

LOCAL Mac - zsh

cd /Users/danb/src/skills
PYTHONDONTWRITEBYTECODE=1 python3 -m unittest tests/test_skillopt_pilot.py

Current skills

Skill Status Purpose
agent-coordination Ready Coordinates multi-vendor AI deliberation through a shared markdown log — preface format, phase model, convergence ledger, approval gate, clean handoff.
handoff Ready Creates a structured session handoff for continuation in a new chat.
grill-me Ready Runs a one-question-at-a-time pressure-test interview for plans/decisions.
parallel-dispatch Ready Generates multi-agent prompts and coordinator playbook from a parallel work plan.
quiz-me Ready Active-recall quiz on a topic, doc, codebase area, or interview prep — one question at a time, adaptive difficulty.
insight-lock Ready Preserves and distills a high-value conversation into durable downstream context — raw capture, distilled insight memo, retrieval hooks, and a light future-use note. The capture skill parallel-dispatch runners use per track.
chat-analysis Ready Analyzes Codex or Claude Code session transcripts for friction patterns, redacts sensitive text, and proposes reviewed agent-instruction improvements.
weekly-update Ready Drafts evidence-driven stakeholder updates from git plus issue-tracker data, with GitHub, Linear, and Jira data contracts.
x-signal-research Ready Researches public X/Twitter conversation signals with Xquik, preserving source ids, query bounds, evidence labels, sample limits, and approval gates for private reads, writes, monitors, webhooks, and bulk jobs.
format-html Ready Multi-skill plugin bundle for HTML formatting + visual safety. Contains the mechanics skills below; format-html/skills/format-html is the overview + installer.
format-html » verify-text-wrap Ready Verifies static HTML portals for caterpillar text, narrow containers, and right-edge layout drift. Canonical home of the wrap-safe runtime library (wrap-safe.css reset + wrapcheck.js probe) it drives.
format-html » table-system-migration Ready Audits, migrates, and regression-tests messy HTML table systems with a public-safe scrub gate.
format-html » number-formats Ready Applies the Macabacus financial number-format standard (en-dash zeros, aligned parenthesized negatives, blue inputs, green links, grey-italic margins, bold totals) to Excel models and HTML tables. Byte-exact codes + openpyxl applicator.
format-html » reskin Ready Detects a repo's design system (a *-design/*-brand skill, tokens stylesheet, or reskin.json) and applies its brand frame (nav/hero/footer + tokens + assets) across the site. Hybrid: runs the repo's bespoke applier if it has one, else a built-in frame injector. Idempotent + dry-run-able.
skill-drift-scanner Ready Audits Codex and Claude skill deployment drift, autosync health, scheduler status, and reload requirements across machine-level skill installs.
blog-image-gen Ready Generates editorial hero images for blog posts through the blog repo's OpenAI image scripts, with current-doc verification, batch/ingest workflows, and thumbnail review.
bogdan-baciu-design Ready Applies Bogdan Baciu's personal editorial design system to sites, prototypes, documents, slide decks, and scoped financial artifacts, bundling tokens, fonts, imagery, previews, and the brand reference.
deep-research-agents Ready Treats Claude Managed Agent, Gemini Deep Research, and Parallel.ai as one deep-research capability with repo-specific harness guidance.

Using these skills

  1. Browse the skill folder.
  2. Open SKILL.md and read the trigger description.
  3. Invoke by naming the skill or matching its trigger phrase in your prompt.
  4. Follow any workflow/output rules exactly.

Tip: keep skills narrowly focused; combine multiple skills only when responsibilities do not overlap.

Authoring and publishing new skills

1) Create a folder

Use a short kebab-case folder name (for example, incident-triage).

2) Add SKILL.md

Start from this minimal skeleton:

---
name: incident-triage
description: Triage production incidents with severity classification and immediate stabilization steps.
---

# Incident Triage

## Purpose
...

## When to use
...

## Workflow
### Step 1 — ...
### Step 2 — ...

## Output format
...

## What this skill does NOT do
...

3) Validate quality

Run through the Skill quality standard.

4) Open a PR

Include:

  • Why the skill is needed.
  • Which user prompts should trigger it.
  • One example session snippet.

Versioning and releases

This repository can be versioned with semantic tags:

  • MAJOR: breaking behavior or format changes to existing skills.
  • MINOR: new skills or backward-compatible expansions.
  • PATCH: typo fixes, clarifications, metadata cleanups.

For production hosting, publish changelog notes so downstream users can pin versions safely.

Contributing

Please read CONTRIBUTING.md for workflow, review expectations, and quality gates.

Security

Please disclose vulnerabilities per SECURITY.md.

License

Distributed under the MIT License. See LICENSE.