菜单
Use when the user asks to forge, scaffold, author, or ship a new skill or agent — the skill-creator's own namesake artifacts. Drives the skill-forge loop (create → fill → validate → critique → ship → publish) via the skill-creator CLI.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | skill-forge |
| description | Use when the user asks to forge, scaffold, author, or ship a new skill or agent — the skill-creator's own namesake artifacts. Drives the skill-forge loop (create → fill → validate → critique → ship → publish) via the skill-creator CLI. |
| description-frequency | on-demand |
| user-invocable | true |
| version | 1.0.0 |
| format | "2025-10-02T00:00:00.000Z" |
| triggers | ["the user asks to create, scaffold, or author a new skill or agent","the user wants to ship or publish a skill through its quality gates"] |
| updated | "2026-07-12T00:00:00.000Z" |
| status | ACTIVE |
Forge a new skill or agent end-to-end using the skill-creator
CLI. The forge is the toolchain that scaffolds, validates, and
ships skills and agents; this skill is the playbook that drives it.
Trigger on any of:
Do not trigger for:
skill-creator refine <name> (corrections, not a new artifact).skill-creator skill retire <name>
(suggest-first, reversible).cartridge-forge skill.docs/SKILL-FORMAT.md is the normative frontmatter/format
reference; docs/skill-authoring/skill-md-quality-checks.md
lists the hard defects the gate rejects; docs/authoring-tools.md
maps the CLI surface. Read them once per session if the context
is not already loaded.name:). Keep it
specific — the description and triggers must not collide with an
existing skill's activation.SKILL.md); an agent is a delegated subagent with its own
tool set (an agents/<name>.md). Author the one the task needs.create → fill → validate → critique → ship → publish
Every step is idempotent. skill ship is the gate — if it exits
non-zero, the skill is not ready to publish.
skill-creator create
The wizard scaffolds a spec-valid SKILL.md: repo-canonical
frontmatter (version / format / status: ACTIVE) plus a body
seeded from a template preset (workflow-guide / reference /
checklist) rather than a blank prompt. It prints a
Next: validate → critique → ship footer.
For an agent instead of a skill:
skill-creator agents create --name=<slug> --description="…"
(or run it interactively). It delegates to the AgentGenerator and
writes a spec-valid agent .md with comma-separated tools.
Replace the template body with real guidance. A well-formed skill body typically wants:
Keep the frontmatter description and triggers distinct: a
trigger that is a verbatim-truncated copy of the description is a
hard defect (see step 3).
skill-creator validate <name>
Runs the skill-content gate: it blocks on structural threats (unsafe YAML, path traversal, schema violation, secret in frontmatter) and sanitizes-and-warns on in-body issues. A clean exit is required.
skill-creator critique <name>
Surfaces quality issues (thin triggers, missing sections, vague description) that validate does not hard-block. Address the findings before shipping.
skill-creator skill ship <name>
Runs the ship pipeline in order — validate → critique →
test-triggering — with a per-gate verdict. Exit code 0 means
<name> is ship-ready. This is the equivalent of cartridge-forge's
eval gate: do not report "done" before it is green. Use --mock
to dry-run the pipeline.
For an agent, the parallel gate is:
skill-creator agents validate
which now exits non-zero on spec defects.
skill-creator publish <name>
Promotes the ship-ready skill. Only run after skill ship is green.
.claude/skills/. That tree is
gitignored and self-mod-guarded; the tracked source is
project-claude/skills/. Author there and let
node project-claude/install.cjs deploy it.skill ship. It is the publish gate.skill-creator refine, not a fresh create.project-claude/…), never
.claude/… — the git-add-blocker enforces this.A well-formed recent skill to model:
project-claude/skills/skill-frontmatter-doctor/SKILL.md — a
single-file, on-demand skill with a tight description, distinct
triggers, and a clear defect-detection procedure.
After the loop completes, report to the user:
name, artifact kind (skill / agent), and source path.version / format / status.validate / critique / ship verdicts (each shown green).install.cjs) and published.Do not report "done" before skill ship (or agents validate) has
been run and shown green.
Runtime HAL for multi-runtime agent orchestration. Detects active AI assistant (Claude Code, Codex, Gemini, Cursor) and exposes uniform interface for startup injection, GUPP enforcement, and communication selection.
Upstream version checking, agentskills.io spec compliance audit, Claude Code feature gap analysis. Activates on updates, version checks, or alignment work.
Use when the user asks "what should I work on", mentions GSD phases, ROADMAP, or .planning/ artifacts. Routes to the correct GSD command for the current project state.
MLOps and the production ML lifecycle -- model packaging and serving, CI/CD for ML, experiment tracking, model registries, reproducibility, production monitoring for data and concept drift, retraining pipelines, A/B and shadow deployment, and rollback. Covers batch vs online/real-time inference, REST endpoints, feature stores, data and version pinning, deterministic pipelines, performance-decay detection, and retraining triggers. Use when deploying models to production, serving predictions, monitoring for data or concept drift, setting up ML CI/CD, tracking experiments, managing a model registry, or planning retraining, shadow rollout, and rollback.
Stellar structure and life cycles from birth to endpoint. Covers the Hertzsprung-Russell diagram, hydrostatic equilibrium, main-sequence physics and the mass-luminosity relation, nuclear fusion stages (pp-chain, CNO cycle, triple-alpha), nucleosynthesis, post-main-sequence evolution through red giants and the horizontal branch, and stellar endpoints — planetary nebulae, white dwarfs and the Chandrasekhar limit, supernovae, neutron stars, and black holes. Use when placing a star on the HR diagram, estimating its lifetime, predicting its fate from its mass, or teaching how a star lives and dies.
Classical control and system dynamics for engineered systems — feedback and closed-loop control, transfer functions and block diagrams, poles/zeros and stability, time-domain response (rise time, overshoot, settling time), PID controllers and tuning, and frequency response (Bode plots, gain and phase margin). Traces feedback control from the flyball governor to modern controllers. Use when analyzing feedback loops, designing or tuning a PID controller, assessing stability from poles or margins, or characterizing transient response of a dynamic system.