chunk-testing-gaps
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes various shell commands including 'docker compose', 'git', and the 'chunk' CLI tool. It also attempts to run arbitrary project-specific scripts found in the repository, such as './do' or 'Taskfile', to perform local triage of code mutations.
- [INDIRECT_PROMPT_INJECTION]: The skill processes project source code and retrieves data from production observability platforms (Honeycomb, Datadog). This surface could be exploited if malicious content in the repository or logs attempts to influence the agent's risk assessment or discovery logic.
- Ingestion points: Project source files, build/test script outputs, and external observability data retrieved via Honeycomb/Datadog integrations.
- Boundary markers: None; the skill does not implement specific delimiters or warnings to isolate instructions from the data being processed.
- Capability inventory: File system modification, shell command execution, and network communication with remote testing sidecars.
- Sanitization: None; data from external sources and the local project is incorporated directly into analysis phases without validation.
- [DYNAMIC_EXECUTION]: The skill modifies production code in-place to create "mutants" and subsequently executes the project's build and test lifecycle scripts to evaluate whether the test suite detects the changes.
- [DATA_EXFILTRATION]: Code fragments and patches are collected into a '.chunk/variants.json' file and transmitted to remote sidecar environments via the 'chunk validate' command to facilitate parallelized testing.
- [EXTERNAL_DOWNLOADS]: The skill utilizes external Docker images, specified via the 'validation.sidecarImage' configuration, to boot throwaway sidecar environments for remote mutation validation.
Audit Metadata