一键在 Manus 中运行任何 Skill

开始使用

add-agent

星标16

分支9

更新时间2026年4月20日 21:15

Scaffold a new Forge subagent (planner/architect/executor/reviewer/debugger/memory-style).

安装

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。

在 Manus 中运行

来源

hoangsonww

hoangsonww/Forge-Agentic-Coding-CLI

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Steps

src/agents/<name>.ts — extend base. Declare:
- allowed tool set (narrow, not "all tools"),
- model preferences (router falls back if unavailable),
- system prompt (short — it's paid for every turn).
Register in src/agents/registry.ts.
Unit test — mock callModel (see test/unit/executor-loop.test.ts). Cover:
- the agent only calls the tools it declared,
- a failing tool call is surfaced as a structured error, not a throw,
- turn caps from src/core/mode-policy.ts are respected.
Docs — update the agent list in docs/ARCHITECTURE.md if user-visible.

Invariants you cannot break:

Do not widen the allowed tool set to "all tools" to make a test pass.
Do not bypass src/core/validation.ts.
Do not call the model directly — always go through router/adapter.

Finish with the verify skill.

同仓库更多 Skills

同仓库

deep-review

hoangsonww/Forge-Agentic-Coding-CLI

Deep review — random exploration + cross-agent review, alternating. Finds integration bugs that self-review misses. Run every 30–60 min during active implementation.

2026-04-2116

flywheel

hoangsonww/Forge-Agentic-Coding-CLI

Overview of Forge's Agentic Coding Flywheel methodology. Use to route between plan, bead, and code space. See FLYWHEEL.md at the repo root.

2026-04-2116

landing

hoangsonww/Forge-Agentic-Coding-CLI

End-of-session ritual. Close finished beads, file new ones for remaining work, run verify, commit in logical groupings, push. Work is not complete until git push succeeds.

2026-04-2116

plan

hoangsonww/Forge-Agentic-Coding-CLI

Start a new Flywheel planning round. Produce initial markdown plan, route to competing frontier models, synthesize hybrid, refine 4–5×. Use before any coding begins on a new feature.

2026-04-2116

plan-to-beads

hoangsonww/Forge-Agentic-Coding-CLI

Translate a polished markdown plan into self-contained beads in .beads/beads.jsonl. Every bead must be executable without reopening the plan. Use after the plan has converged.

2026-04-2116

polish-beads

hoangsonww/Forge-Agentic-Coding-CLI

Polishing pass over all open beads. Run 4–6 times. "Check your beads N times, implement once." Use after plan-to-beads and between refinement rounds.

2026-04-2116

name	add-agent
description	Scaffold a new Forge subagent (planner/architect/executor/reviewer/debugger/memory-style).

Agents extend src/agents/base.ts. Copy from the simplest existing agent (src/agents/reviewer.ts or src/agents/debugger.ts).

Steps

src/agents/<name>.ts — extend base. Declare:
- allowed tool set (narrow, not "all tools"),
- model preferences (router falls back if unavailable),
- system prompt (short — it's paid for every turn).
Register in src/agents/registry.ts.
Unit test — mock callModel (see test/unit/executor-loop.test.ts). Cover:
- the agent only calls the tools it declared,
- a failing tool call is surfaced as a structured error, not a throw,
- turn caps from src/core/mode-policy.ts are respected.
Docs — update the agent list in docs/ARCHITECTURE.md if user-visible.

Invariants you cannot break:

Do not widen the allowed tool set to "all tools" to make a test pass.
Do not bypass src/core/validation.ts.
Do not call the model directly — always go through router/adapter.

Finish with the verify skill.