- Git
- GitHub CLI (
gh) — used by skills for all GitHub operations - A compatible coding agent that supports Agent Skills slash commands
No language runtime, package manager, or build tools are needed. The repository contains only Markdown files.
git clone <repo-url>
cd forgeNo dependency installation required.
- Create a directory under
skills/namedforge-<skill-name>/ - Create a
SKILL.mdfile inside it - Add YAML frontmatter with
nameanddescription(see Architecture for field reference) - Write the structured prompt body following the section order: Title → Input → Process → Guidelines → Related Skills → Example Usage
- Update workflow references in relevant skills' "Related Skills" sections
- Update the docs table in
AGENTS.mdandREADME.mdif a new doc category is needed
- Read the full SKILL.md to understand the current flow
- Make targeted changes — avoid rewriting entire skills when a small edit suffices
- If changing conventions (commit format, branch naming, guidance filenames, structured-input syntax, etc.), grep across all relevant skills to update consistently:
grep -r "conventional commit" skills/ grep -r "branch naming" skills/ grep -r "AGENTS.md\|CLAUDE.md" skills/ grep -r "additional context\|-- <additional context>" \ skills/forge-setup-project \ skills/forge-brainstorm \ skills/forge-implement \ skills/forge-reflect \ skills/forge-address-pr-feedback
- Verify cross-skill consistency: shared conventions must be identical in every skill that references them
- If adding or modifying a
(delegate)step, ensure sub-agent instructions are self-contained (or composed from a role reference + task instructions) and the skill works correctly when executed inline - If modifying a role file, verify all skills referencing that role still work correctly
Roles are sub-agent persona definitions that live inside the skill directory that uses them (under roles/).
- Create a
roles/directory inside the skill that will use it - Create a
<role-name>.mdfile there - Add YAML frontmatter with
name,description, and optionallymodel-hint - Write the prompt body: identity, behavior rules, output format, constraints
- Update the skill’s delegation step to reference the role
Extract a persona into a role file when:
- The persona has calibration rules or behavior that benefits from separation (severity rubrics, research methodology)
- The skill body is approaching the instruction budget and the persona can be loaded on demand
Don’t extract trivial personas that are just a sentence or two — inline them in the skill’s blockquote.
If multiple skills need the same role, duplicate the file into each skill. Self-containment beats DRY for distributed prompt files.
- Read the role file and the skill that uses it
- Make targeted changes
- Verify the skill’s delegation step still works with the updated role
- Check the
model-hintis still appropriate - If the same role is duplicated across skills, update all copies
This is a documentation-only repository. The relevant commands are:
| Command | Purpose |
|---|---|
git status |
Check working tree state |
git diff |
Review changes before committing |
gh issue list |
List open issues |
gh pr list |
List open pull requests |
gh issue create |
Create a new issue |
gh pr create |
Create a pull request |
Before committing changes to any SKILL.md:
- Cross-reference check: Ensure conventions mentioned in the modified skill match all other relevant skills (commit format, branch naming, workflow order, canonical guidance file, shared trailing context syntax where applicable)
- Workflow continuity: Verify the "Related Skills" section correctly links to the next relevant step
- Frontmatter validity: Confirm
nameanddescriptionare present and accurate - Bash example accuracy: All
ghandgitcommands in examples must be valid and runnable