devenv-project
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to read and interpret
devenv.nixanddevenv.yamlfiles, which contain shell commands and configuration options. Maliciously crafted project files could attempt to influence the agent's behavior through instructions embedded in comments or code blocks. - Ingestion points: The agent reads
devenv.nix,devenv.yaml, anddevenv.lockfrom the local project directory. - Boundary markers: The skill provides high-level warnings to the agent (e.g., "Do not silently grant trust"), but the files being processed do not contain inherent safety boundaries.
- Capability inventory: The skill documentation describes using the
devenvCLI to perform tasks that execute shell code, such asdevenv shell,devenv up, anddevenv test(referenced inSKILL.mdandreferences/workflows.md). - Sanitization: There is no automated sanitization of the input files; the agent is expected to manually inspect content as instructed in
SKILL.md. - [COMMAND_EXECUTION]: The skill provides instructions for defining and executing shell scripts, tasks, and processes within the development environment.
- Evidence:
references/workflows.mdsections 8-14 describe how to definescripts,tasks, andprocessescontaining arbitrary bash code (e.g.,scripts.silly-example.exec = 'curl ... | jq ...'). - Evidence:
SKILL.mdinstructs the agent to choose validation according to side effects, noting thatdevenv shell,tasks,devenv up, anddevenv testall execute code defined in the configuration. - [EXTERNAL_DOWNLOADS]: The skill manages environments that rely on external dependencies and remote scripts.
- Evidence:
references/workflows.mdsection 6 describes fetching a configuration script fromhttps://raw.githubusercontent.com/cachix/devenv/82c0147677e510b247d8b9165c54f73d32dfd899/direnvrcusingsource_url. - Evidence:
references/configuration.mdsection 4 describes usingdevenv.yamlto specify remote inputs from GitHub (e.g.,github:owner/repo). - Note: These references target well-known services and organizations, which is common for environment management tools.
Audit Metadata