1# Forge
 2
 3Research-backed skills and configs for LLM agent tools.
 4
 5## Critical Rules
 6
 7- ALWAYS add new skills to the Available Skills table in README.md when creating them
 8- ALWAYS follow the SKILL.md frontmatter spec (name, description, optional: disable-model-invocation, argument-hint)
 9- NEVER add co-author tags to commits
10- NEVER commit without explicit ask
11
12## Architecture
13
14- `skills/` - each skill is a directory with `SKILL.md` following the Agent Skills open standard
15- Skills are portable across Claude Code, Cursor, Copilot, Codex, Crush, and any SKILL.md-compatible agent
16- Each skill is self-contained - no cross-skill dependencies
17
18## Stack Decisions (Locked)
19
20- Skills use SKILL.md format (YAML frontmatter + markdown body)
21- No build step, no compilation, pure markdown
22- MIT licensed
23
24## Commands
25
26No build system. Skills are markdown files.
27
28```
29# Test a skill locally
30cp -r skills/<name> ~/.claude/skills/
31
32# Symlink for dev
33ln -s $(pwd)/skills/<name> ~/.claude/skills/<name>
34```
35
36## Implementation Pitfalls
37
38- Skill descriptions over 60 words waste token budget
39- `disable-model-invocation: true` removes description from context entirely - don't over-optimize descriptions for these
40- Skill body over 500 lines should be split into supporting files
41- Don't duplicate content between description and body
42
43## Commit Style
44
45Conventional commits: `type(scope): description`
46
47NEVER add co-author tags.
48
49## Compact Instructions
50
51Always keep: current skill being edited, research findings being applied, README.md update status.
52
53## Do NOT
54
55- Create skills without adding them to README.md Available Skills table
56- Write descriptions longer than 60 words
57- Add Claude-specific language when the skill works across agents
58- Use long dashes
59- Add AI slop phrases