Asgard Skills — Repository Context

This repository is the raw ingredient pantry for Asgard AI Platform plugins: 263 coding agent skills organized by topic prefix.

Repository Purpose

• NOT a plugin: this repo holds raw skill files; plugins are assembled elsewhere by combining curated skills + MCPs
• NOT a template: see skill-template for the plugin scaffolding
• IS a versioned library of methodology + judgment + gotchas, packaged as Claude Agent Skills

Skill Anatomy

Every skill is a directory at the repo root, named {category}-{slug}, containing:

{category}-{slug}/
├── SKILL.md           ← REQUIRED: frontmatter + methodology body
├── examples/          ← REQUIRED: directory must exist (uses .gitignore trick for empty dirs)
├── references/        ← REQUIRED: directory must exist; populated when SKILL.md offloads heavy content
└── scripts/           ← OPTIONAL: only present when the skill has executable scripts

Naming Convention

Skills use topic-based prefixes (no skill- prefix). The 22 categories:

grad-      Graduate-level theoretical models
algo-      Algorithms
biz-       Business school frameworks
hum-       Humanities / critical reasoning
tw-        Taiwan-specific knowledge
ecom-      E-commerce practical
econ-      Economics fundamentals
meta-      Interdisciplinary mental models
ops-       Business operations
law-       Legal frameworks
pr-        PR & brand communications
cs-        Customer service
data-      Data analytics
mfg-       Manufacturing
mkt-       Digital marketing
med-       Mass communication / journalism
soc-       Social science
stat-      Statistical methodology
tech-      General tech (API, prompt eng, MCP)
ux-        Design / UX methodology
fin-       Finance practical
xborder-   Cross-border commerce

SKILL.md Rules

1. Frontmatter (name, description required; metadata optional)
2. Description: WHAT + WHEN, imperative tone, < 1024 chars, no XML angle brackets, includes trigger phrases
3. Body: < 500 lines / < 5,000 tokens — offload heavy content to references/
4. IRON LAW block: every skill must declare one non-obvious constraint
5. Output Format: a Markdown or JSON template the agent should follow
6. Gotchas: 5-6 specific, actionable pitfalls (not generic advice)
7. Skill name must not contain: claude or anthropic (reserved)

Design Patterns by Section

Scripts (Deterministic Calculators)

20 skills ship pure-stdlib Python scripts. Standards:

• Single file in {skill}/scripts/
• --help, --input <json>, --verify (self-test) flags
• Outputs JSON to stdout
• Pure stdlib (no numpy/scipy/etc. unless documented)
• Errors as JSON to stderr
• Built-in verify() function with assertions on known input/output pairs

When adding a new script:

1. Implement compute() with type-checked inputs
2. Implement verify() with at least 3 assertions covering happy path, edge case, and error case
3. Wire up main() with argparse
4. Update the SKILL.md ## Scripts section with usage example

Quality Audit Anti-Patterns to Avoid

When writing or modifying a skill, watch for:

1. Phantom Trigger — description over-matches (e.g., "everything is connected" triggers on any vague reference)
2. Kitchen Sink — SKILL.md crammed with everything; should offload to references/
3. Over-Teaching — explaining basics the agent already knows (regression, p-values, prompt engineering)
4. Missing Gotchas — generic advice instead of specific traps the agent would fall into
5. Procedural Straitjacket — too rigid where judgment is needed; or too vague where precision matters
6. Truism Iron Law — Iron Law that any competent agent would already follow ("be specific", "look at the data first")
7. Non-verifiable Sample I/O — algo skills must have exact computable input → output, not approximate ranges
8. Broken references/ pointers — SKILL.md references files that don't exist

Working in This Repo

• Test scripts after editing: python <skill>/scripts/<script>.py --verify
• Cross-skill references should use the new naming (no skill- prefix)
• The repository is published under MIT license; do not check in proprietary content
• See TODO.md for current work in progress and pending items

See Also

• skill-template — Plugin template repo
• Asgard MCPs — Data-provider MCPs
• Claude Agent Skills documentation