一键导入
mordred-status
Read-only inspection of the Mordred privacy layer (status, policy, network path, plugins). Observe only — never mutate or read secrets.
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。
菜单
Read-only inspection of the Mordred privacy layer (status, policy, network path, plugins). Observe only — never mutate or read secrets.
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
Run the mordred-hermes release runbook — version bump, dev PR, dev→main release PR, TestPyPI dry run, verification, PyPI publish, tag. Use when the user asks to release, publish, or bump the package version.
Synthetic test fixture — a skill that declares Tor as its required network path.
Synthetic test fixture — a skill that opts into Mordred keyvault protection.
Synthetic test fixture — a skill that declares clearnet network access.
Synthetic test fixture — a skill with no metadata.mordred extension at all.
| name | mordred-status |
| description | Read-only inspection of the Mordred privacy layer (status, policy, network path, plugins). Observe only — never mutate or read secrets. |
| version | 1.0.0 |
| metadata | {"hermes":{"tags":["mordred","privacy","status","policy","network","keyvault","encryption"],"related_skills":[]},"mordred":{"network_requirements":"local-only","requires_keyvault":false}} |
Inspect the local Mordred privacy layer when the user asks things like "mordredの状態は?" / "what's my mordred status" / "今のpolicyは?" / "どのnetwork経路?" / "mordredのプラグイン一覧".
Mordred's job is to constrain and audit this agent. You may observe its state but you must never change it or read its secrets from a conversation.
ALLOWED (metadata only — no secrets):
| Intent | Command |
|---|---|
| Overall state | MORDRED status |
| Machine-readable state | MORDRED status --json |
| Active policy | MORDRED policy show |
| Explain a skill's install decision | MORDRED policy explain <skill-id> |
| Network path + liveness | MORDRED network status |
| Discovered Mordred plugins | MORDRED plugins list |
where MORDRED is resolved by trying, in order (the CLI's real location is
environment-specific — never hardcode one developer's machine path):
command -v hermes-mordred — use it if it resolves on PATH.README.md ("Install (users, from PyPI)" section):
~/.hermes/hermes-agent/venv/bin/hermes-mordred.mordred-hermes venv
(e.g. "I can't find hermes-mordred on PATH or at the default Hermes venv
location — what path did you install it to?"). Only conclude "not
installed" (see below) once the operator confirms there is no such install.Run these via the code-execution / shell tool and report the output.
FORBIDDEN — never run from this skill. If the user wants any of these, print the exact command and tell them to run it themselves in their shell:
configure, upgrade, keyvault init|recover|reset|enable-se|enable-tpm,
encryption enable|disable|purge, network init|use, vault *, install.audit decrypt, vault cat.audit tail|grep — it can surface install decisions into the
transcript. Do not run it; hand the command to the user instead.These rules preserve the "domain separation" principle in
docs/dev/HARNESS_PRIVACY.md: a recording agent may watch Mordred's
on/off state, but secrets (recovery passphrase, .env contents, decrypted audit)
and control actions must stay on the operator's shell, outside the transcript.
Residual risk: the FORBIDDEN list above is a prompt-level instruction, not an enforcement mechanism — it does not technically prevent this agent from running a mutating command. Hard enforcement of what the agent may execute belongs to the Hermes exec-tool approval settings, not to this skill (see
README.md§ "Conversational read-only access").
If none of the three resolution steps above find hermes-mordred —
including after asking the operator — tell the user no Mordred install was
found and stop. Do not try to install or build anything.