Policy pack · Developer
Claude Code safety policy pack
Default safety manifest for Claude Code agents. Every policy addresses a real risk that developers face when using autonomous coding agents: credential exfiltration, destructive commands, scope escape, runaway and runaway loops.
What's in it
13 policies, 13 of them blocking.
Critical policies block the step before it runs. High, medium and low ones let it run and open an incident.
| Policy | Checked | Severity | Rule |
|---|---|---|---|
| No exec after secret read | Before every step | critical | tainted_path_block |
| No network after secret read | Before every step | critical | tainted_path_block |
| No force push | Before every step | critical | not |
| No recursive delete of root | Before every step | critical | not |
| No recursive delete of home | Before every step | critical | not |
| No recursive delete via parent traversal | Before every step | critical | not |
| No recursive delete of system directories | Before every step | critical | not |
| No git reset --hard | Before every step | critical | not |
| No write outside project | Before every step | critical | all_of |
| No shell write to sensitive paths | Before every step | critical | not |
| No shell read of secret files | Before every step | critical | not |
| Max exec per task | Before every step | critical | execution_max_steps |
| Max consecutive exec | Before every step | critical | max_consecutive_same_type |
Version 1.0.5. Source: manifests/developer/claude-code-safety.yaml.
Assign it
One command, no redeploy.
Assign it from the dashboard or the CLI. Running agents pick up an assignment within the policy refresh window, five minutes by default.
- Policies
- 13, of which 13 are checked before every step.
- Version
- 1.0.5
$ kyvvu list-manifests$ kyvvu assign-manifest --agent-id <agent> --repo-id <repo> \ --manifest manifests/developer/claude-code-safety.yamlAbout this pack
All runtime policies are severity: critical (= block). These are genuinely dangerous actions where "warn but allow" makes no sense. Teams can downgrade individual policies to high (= warn) if they prefer softer enforcement.
Designed for agents registered via kyvvu-claude. Assign via the Kyvvu dashboard or API.
Runtime policies in this pack
- No exec after secret read (critical)
- No network after secret read (critical)
- No force push (critical)
- No recursive delete of root (critical)
- No recursive delete of home (critical)
- No recursive delete via parent traversal (critical)
- No recursive delete of system directories (critical)
- No git reset --hard (critical)
- No write outside project (critical)
- No shell write to sensitive paths (critical)
- No shell read of secret files (critical)
- Max exec per task (critical)
- Max consecutive exec (critical)
From Kyvvu's open policy library at github.com/Kyvvu/manifests, Apache 2.0.
Next
Run the whole pack in your browser.
In your browser
Try it in the playground
The policy is loaded. Add the steps your agent takes and watch the engine decide, step by step.
On your machine
Five minutes, no account
kyvvu try runs a demo agent against the bundled baseline. Then assign this pack from the dashboard or the CLI.
For security teams
See it on your agents
A 30-minute demo on the agents you actually run, with the policies that fit them.