ck
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The
/ck:initcommand parses local project files to auto-detect context, which could contain malicious instructions. - Ingestion points: The
commands/init.mjsscript readsCLAUDE.md,README.md,package.json, and.git/configfrom the current working directory. - Boundary markers: Extracted content from sections like "Current Goal" or "Do Not Do" is presented to the agent for confirmation, but it lacks explicit protective delimiters or instructions to ignore embedded commands.
- Capability inventory: The skill can read and write files in the
~/.claude/ck/directory and executegitcommands viaspawnSync. - Sanitization: The script performs regex-based extraction but does not sanitize the resulting text against prompt injection payloads that might target the agent during the initialization process.
- [COMMAND_EXECUTION]: The skill's instructions in
SKILL.mddirect the agent to interpolate its own generated JSON into a shell command string. - Evidence: The
/ck:saveand/ck:initcommands use the patternecho '<json>' | node .... This creates a risk of command injection if the agent-generated JSON contains unescaped single quotes or other shell-active characters. - Evidence: The skill uses
child_process.spawnSyncto executegitcommands inshared.mjsandhooks/session-start.mjs. While it uses the array form of arguments which mitigates most injection, it relies on parameters likesinceDateandprojectPathstored in local JSON configuration files.
Audit Metadata