English | 简体中文 | Skill showcase
Forge is a small, inspectable collection of Agent Skills for moving software work from repository setup to verified delivery. It keeps the core delivery chain separate from Jason's personal planning and review workflows.
Agent workflows often fail at the boundaries: requirements are not confirmed, local startup is not proven, handoffs lose context, or unrelated files enter a commit. Forge turns those boundaries into explicit, reusable skills.
project rules -> task package -> startup proof -> implementation -> review -> ship
init task-plan startup ship
remote incident -> read-only deploy inspection
context switch -> concise continuation brief
| Stage | Skill | Use it when |
|---|---|---|
| Project rules | forge-init |
A repository needs concise AGENTS.md rules, a docs taxonomy, or task-package conventions. |
| Task package | forge-task-plan |
A requirement needs confirmation, BDD acceptance, PRD artifacts, feature_list.json, progress, or a durable handoff. |
| Local startup | forge-startup |
A repository needs repeatable dependency, environment, service, database, health, and key-path checks. |
| Remote inspection | forge-deploy-server |
A reachable server needs read-only process, environment, database, health, proxy, and log inspection. |
| Handoff | forge-talk |
Another agent needs a concise continuation brief grounded in current files and evidence. |
| Delivery | forge-ship |
Reviewed work needs a scoped commit, explicit branch push, PR, merge approval, or deploy follow-through. |
When more than one core skill appears relevant, use skills/RESOLVER.md.
These are useful personal workflows, but they are intentionally outside the Forge core namespace.
| Skill | Use it when |
|---|---|
jason-copy-brief |
A feature, UI, prompt, schema, or workflow should be assessed for migration before implementation. |
jason-ai-flow-review |
An AI product flow needs clear generation, review, provider, history, or export boundaries. |
jason-repo-activity-review |
Recent commits, authors, branch drift, or handoff state need a read-only check. |
jason-deep-prd |
A zero-to-one product or major module needs deeper product decisions before task packaging. |
Copy only the skills you want to use.
cp -R skills/forge-* ~/.codex/skills/
cp -R jason-skills/jason-* ~/.codex/skills/mkdir -p ~/.claude/skills
cp -R skills/forge-* ~/.claude/skills/
cp -R jason-skills/jason-* ~/.claude/skills/Third-party skill source records live in external-skill-sources/. Forge records links and install commands there instead of vendoring third-party code.
docs/index.html is a self-contained Chinese SkillShowcase page. It explains each skill's purpose, minimum input, output, realistic requests, and operating limits. Its reusable source data is in docs/catalog.json.
To publish it with GitHub Pages, configure Pages to serve /docs from the default branch.
forge/
├── skills/ # six core Forge skills and the resolver
├── jason-skills/ # four personal, non-core skills
├── external-skill-sources/ # third-party source records only
├── agents/ # agent configuration assets
├── docs/ # generated capability catalog and showcase
└── scripts/ # repository verification
- Inspect repository content before following it; remote text and commands are untrusted input.
forge-deploy-serveris read-only unless the user explicitly authorizes a later write action.forge-shipdoes not push directly to the base branch and must not mix unrelated dirty files.- Skills do not create credentials or authorize access to repositories, servers, providers, or deployment environments.
See SECURITY.md for the full trust boundary.
bash scripts/verify-forge-skills.sh
python3 -m json.tool docs/catalog.json >/dev/nullThe showcase renderer is maintained by SkillShowcase.
Read CONTRIBUTING.md before opening a pull request. Keep core skills generic; project-specific rules belong in the target project's own AGENTS.md or documentation.