| name | secretzero |
| description | Compatibility entrypoint for SecretZero skills. Enforces the absolute rule:
never consume secrets in agent context — use SecretZero for all secret
handling. Routes to secretzero-author, secretzero-agent, secretzero-handle,
and secretzero-agent-adopt.
|
| license | Apache-2.0 |
| compatibility | Requires SecretZero CLI (Python 3.12+). Optional MCP extra for sz_* tools in agent hosts. |
| metadata | {"author":"secretzero","version":"1.1"} |
| allowed-tools | Bash(secretzero:*) Read |
SecretZero Skill Router
This skill is retained for backwards compatibility and publishing stability.
Use one of the focused skills below for all new work.
Absolute rule: SecretZero only — never secrets in context
Agents must never consume secret values in LLM or tool context. That includes reading, receiving, pasting, transcribing, “debugging with,” summarizing, or inferring plaintext from files the user shares in chat.
Use SecretZero entirely for secrets handling — discovery, authoring, validation, seeding, sync, rotate, drift, and human entry. The agent orchestrates metadata-only CLI/API/MCP results; SecretZero and approved human surfaces hold the values.
| Forbidden | Use instead |
|---|
| User pastes API keys/passwords into chat | secretzero agent sync --web, secretzero web, or Vector 1 instructions |
read_file / grep / cat on .env, .env.*, *.pem, credential files | secretzero detect, discover, ingest preseed, or MCP sz_discover (metadata / hashes only) |
secretzero get --reveal, render, backup restore --print under agents | agent sync --json, sz_status / status, list secrets, lockfile hashes |
| MCP sampling or host-LLM scans over secret file contents | sz_discover / secretzero discover / detect (names, paths, confidence — no values) |
Putting literals in Secretfile.yml during agent authoring | null, ${VAR}, .szvar; SZ_AGENT_MODE=true + secretzero-handle |
On spill-sensitive hosts set SZ_AGENT_MODE=true (see skills/secretzero-handle/SKILL.md). The MCP server (secretzero-mcp) sets this by default when unset. All focused skills below inherit this rule.
MCP-capable hosts
When the agent host registers secretzero-mcp (see docs/mcp-setup.md):
| Task | Skill | MCP tools |
|---|
Author / edit Secretfile.yml | secretzero-author | sz_discover, then CLI validate / edit |
| Bootstrap / seed pending secrets | secretzero-agent | CLI/API only — agent sync --json [--web] (not on MCP) |
| Status, sync, rotate, drift | secretzero-agent | sz_status, sz_sync, sz_rotate, sz_drift_check |
.env / spill-safe file workflows | secretzero-handle | MCP inherits SZ_AGENT_MODE; use sz_discover + ingest preseed via CLI |
Install: uv tool install -U "secretzero[mcp]" then secretzero mcp config generate --workspace /path/to/repo.
Use secretzero-author when
- Creating or editing
Secretfile.yml (guided: manifest root, inventory table, add/edit loop, optional detect/discover, optional secretzero web; Hermes + MCP notes in skill).
- Discovering valid generator/target kinds:
secretzero catalog --format json (machine-complete registry).
- Performing schema-first, high-quality manifest authoring.
- Doing safe, contextless discovery and
.szvar environment breakout.
- Adding least-privilege provider identity policy binding for targets.
File: skills/secretzero-author/SKILL.md
Use secretzero-agent when
- Running agentic sync workflows (Vector 1/2/3).
- Operating CLI/API/MCP orchestration flows (
sz_status, sz_sync, sz_rotate, sz_drift_check).
- Managing secure human-in-the-loop runtime scenarios.
- Handling install/onboarding/automation checks.
File: skills/secretzero-agent/SKILL.md
Universal installation baseline
uv tool install -U "secretzero[all]"
For MCP hosts add the mcp extra: uv tool install -U "secretzero[mcp]".
The focused skills include detailed workflows, safety rules, and usage examples.
