メニュー
Run agents through a lightweight `agent-browser`-style CLI. Use when another agent needs `agent-run codex|cursor|claude-code|ai-sdk ...` as an immediate direct call, or local send/peek/wait sessions without depending on the daemon.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use go/setup/audit/goal/design/fact/reframe/close as the stable entrypoint for ordinary agent work. Use when starting or resuming work, choosing autonomous action vs review vs escalation, changing system shape, recording progress, verifying claims, or closing work with a durable record.
Methodology for writing argumentative technical articles — pieces that convince the reader of a specific claim through a chain of reasoning. Use when the user wants a technical article, essay, or book chapter that takes a position — introducing a new idea, challenging an accepted one, or arguing for a particular approach. Triggers on phrases like "write an article/essay/chapter about X," "make the case for X," "argue that Y," or structuring a multi-chapter argument about agents, architectures, systems design, or methodology. Do NOT trigger for tutorials ("how to do X"), reference or API documentation, release notes, changelogs, or pure explainers where no claim is being argued — those need a different framework.
Refactoring existing articles or drafts using a code-refactoring metaphor — extract information units (facts, claims, examples, metaphors) into a neutral inventory, rewrite the skeleton for a new thesis or audience, then inline units with expansion, compression, or transformation. Triggers on "重构这篇文章", "把这几篇合并改写", "用这些素材重写一篇", "这篇骨架我想换但素材想保留", "把旧博客整理重写", "refactor this article", "restructure this draft". Use to restructure existing pieces, merge drafts, repurpose for a different audience, or salvage old writing. Do NOT trigger for writing from scratch (use technical-article-writing) or light polish (use technical-article-writing's `revise`).
Writer voice profile — an MBTI-inspired adaptive assessment that captures a person's cross-genre writing taste and style preferences. Use when the user wants to build or update their writing profile, calibrate AI writing tools to their voice, or find out "what kind of writer am I". Triggers on phrases like "写作画像", "writing profile", "writer type", "我是什么类型的作者", "测一下我的写作风格偏好", "calibrate my writing voice", or any explicit mention of building a writer profile. Produces a file at `~/.claude/writing-profile/profile.md` that downstream writing skills (technical-article-writing, future blog/diary skills) read to adapt AI output to the user's voice.
Run comparative development evaluations: assign the same task to agent-worker team AND claude-code (team subagents), then have an evaluator agent score both and a reviewer agent verify the evaluation. Fully orchestrated — you coordinate all phases, agents notify via channel/callback when done. Use this skill when the user wants to benchmark agent-worker vs claude-code, run a dev eval, compare agent performance, or prove that the workspace team outperforms solo agents. Trigger on phrases like 'dev eval', 'run eval', 'compare performance', 'benchmark', '对比评估', '开发评估', '跑评估', 'eval iteration', '/dev-eval'.
Run web UI performance tests for the agent-worker project. Automates build, daemon restart, cache-bust version bump, Lighthouse audit, and metric reporting — all in one command. Use this skill when the user says "run perf", "check performance", "lighthouse", "perf test", "why is it slow", "check TBT", "性能测试", "跑一下性能", or wants to measure/compare web UI performance. Also trigger after making changes to semajsx/style, render, or any web UI component when performance impact needs verification.
| name | agent-run |
| description | Run agents through a lightweight `agent-browser`-style CLI. Use when another agent needs `agent-run codex|cursor|claude-code|ai-sdk ...` as an immediate direct call, or local send/peek/wait sessions without depending on the daemon. |
Use this skill when another agent needs a small command-line surface for invoking an agent-worker runtime. The default experience should feel like agent-browser: one executable, one short command, direct output, local sessions, no daemon ceremony.
Read references/runtime-wrapper.md when changing the CLI or adding a backend.
This skill is the lightweight CLI side of agent-worker's headless-first product shape. Use it for shell-capable agents that need an immediate invocation surface or a small set of local background sessions. Use the daemon or MCP surface only when the caller needs shared harness state, durable task ledgers, multi-agent coordination, or typed scheduling.
skills/agent-run/scripts/agent-run codex "Investigate this failure and report the smallest fix."
skills/agent-run/scripts/agent-run cursor "Review the current diff and report risks."
skills/agent-run/scripts/agent-run ai-sdk "Summarize this repository state."
skills/agent-run/scripts/agent-run codex "Implement this bounded envelope" --worker-rule attention-driven
skills/agent-run/scripts/agent-run codex send "Fix the bug" --wait 5m
skills/agent-run/scripts/agent-run codex send "Explore auth failure" --session auth-failure --no-wait
skills/agent-run/scripts/agent-run sessions
skills/agent-run/scripts/agent-run sessions clean --older-than 7d
skills/agent-run/scripts/agent-run codex wait --wait 2m
skills/agent-run/scripts/agent-run codex peek --session auth-failure
skills/agent-run/scripts/agent-run codex status
Use --json when another agent or script will parse the result.
Use --skill <name> to load a project skill and prepend it to the agent's message (searches skills/<name>/SKILL.md in the local project, then ~/.hermes/skills/).
Use --worker-rule <name> for delegated worker prompts. It loads only skills/<name>/references/worker.md, not the full SKILL.md.
Use --timeout <duration> (default 300s) to cap one-shot invocations.
The CLI mirrors the project interaction model:
| Command | Meaning |
|---|---|
agent-run <agent> "message" | direct one-shot invocation, daemonless |
agent-run <agent> "message" --skill <name> | one-shot with a project skill inlined |
agent-run <agent> "message" --worker-rule <name> | one-shot with only worker-facing rules inlined |
agent-run <agent> send "message" | direct send, optionally starting a local session |
agent-run <agent> wait | wait on a local session |
agent-run <agent> peek | inspect local session output |
agent-run <agent> status | direct runtime preflight |
agent-run sessions | list tracked background sessions |
agent-run sessions clean | remove completed/failed sessions older than 7d |
Options:
--model <model>: override backend model. For ai-sdk, use provider:model; agent-run checks the API key for that provider. If omitted, it scans local API key env vars and picks the first usable default, with DEEPSEEK_API_KEY first.--skill <name>: load a skill from skills/<name>/SKILL.md and prepend it to the task. Searches project-local skills/ first, then ~/.hermes/skills/.--worker-rule <name>: load skills/<name>/references/worker.md and prepend it to the task. Use this for bounded delegated workers, especially --worker-rule attention-driven.--timeout <duration>: cap one-shot invocations (default 300s). Use 30s, 5m, 1h.--wait <duration>: wait duration such as 30s, 5m, 1h.--session <name>: label a background run so parallel work can be tracked separately.--no-wait: send only.--older-than <duration>: for sessions clean, remove sessions older than this (default 7d).Prefer the highest existing layer that fits the caller:
| Caller need | Use |
|---|---|
| Another agent wants a simple immediate command | skills/agent-run/scripts/agent-run <agent> "..." |
| Another agent wants local background / waitable work | skills/agent-run/scripts/agent-run <agent> send ... --session ... --no-wait |
| Another harness/agent wants durable task delegation | task_create + task_dispatch(worker: "<agent>"), or aw task new + aw task dispatch --to <agent> |
| Repo code needs a runtime directly | AgentLoop implementations from @agent-worker/loop |
| Runtime selection comes from harness config | packages/agent-worker/src/loop-factory.ts |
| You need to inspect protocol drift | backend-specific wrapper, such as internals/loop/src/loops/codex.ts |
Do not force daemon startup for one-shot or local session paths. Use direct runtime access there. Do not recreate harness/task-ledger features inside agent-run; those remain daemon/MCP responsibilities.
Direct runtime loops do not automatically load this repo's skills/ directory. Use --skill <name> to explicitly load a skill — agent-run reads the SKILL.md file and prepends it to the agent's message. Searches project-local skills/<name>/SKILL.md first, then ~/.hermes/skills/.
Do not use --skill attention-driven for a worker. attention-driven/SKILL.md
is Main-Agent-only. Use --worker-rule attention-driven so the worker receives
only references/worker.md plus the bounded task envelope.
Supported direct backends are codex, cursor, claude-code / claude, and ai-sdk / ai. claude-code is wired for parity, but do not use it in this workspace while the Claude Code subscription is paused.
For parallel work, start each task with a session label:
skills/agent-run/scripts/agent-run codex send "Audit auth flow" --session auth-audit --no-wait
skills/agent-run/scripts/agent-run cursor send "Check build failure" --session build-fail --no-wait
skills/agent-run/scripts/agent-run sessions
skills/agent-run/scripts/agent-run codex wait --session auth-audit --wait 5m
skills/agent-run/scripts/agent-run codex peek --session build-fail
This borrows the useful part of Claude Code's agent view interaction: keep a compact list of background sessions and jump back only when one needs attention or has output.
ACP remains a useful mental model for the user-facing contract:
| ACP idea | agent-run / agent-worker surface |
|---|---|
| one-shot prompt | agent-run <agent> "..." |
| session prompt | agent-run <agent> send ... |
| session update | agent-run <agent> wait ... |
| session history | agent-run <agent> peek ... |
| initialization/readiness | agent-run <agent> status |
| runtime implementation | AgentLoop / backend wrapper |
For ordinary use, stay at the CLI. Read references/runtime-wrapper.md only when extending or debugging the implementation.