Policy library#
Built-in policies are starting points, not compliance attestations. Choose a baseline for the runtime you own, inspect the resolved result, and add tests for your allowed and forbidden effects.
Built-in rulesets#
| Name | Starting posture |
|---|---|
default | General-purpose baseline |
strict | More restrictive baseline |
permissive | Broad permissions for controlled development |
ai-agent | Agent-oriented rules |
cicd | CI/CD-oriented rules |
remote-desktop | Computer-use and desktop rules |
panic | Emergency deny policy |
Read the exact ruleset sources before selecting one. Inheritance replaces supplied core rule blocks; it does not append to their arrays.
hushspec: "1.0.0"
name: repository-agent
extends: builtin:ai-agent
rules:
tool_access:
allow: [read_file, search]
require_confirmation: [write_file]
block: [deploy]
default: blockInspect and test#
h2h resolve policy.yaml
h2h validate --strict policy.yaml
h2h test policy.test.yamlUse the quickstart test suite as the shape for your test file. Review the full resolved policy whenever replacing a baseline block. A child can intentionally weaken a base during inheritance; origin projection is the separate mechanism that only narrows permissions.
Domain-oriented templates#
The builtin:library/<vertical>/<name> namespace includes general,
healthcare, finance, government, education, and DevOps templates.
For example, builtin:library/general/recommended is a general starting point.
See the library catalog
for names and rationale. Labels such as HIPAA or SOC 2 identify intended
control mappings, not a legal conclusion or audit opinion.
Built-in policy bytes can retain a supported 0.x version. Do not rewrite those
bytes merely for visual consistency: doing so changes hashes and evidence.
Your new authored leaf should use 1.0.0.