The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run-store.py executes git commands locally using subprocess.run. The implementation uses a safe argument list structure, and all variable inputs (such as the repository path and run ID) are strictly validated for path traversal patterns and restricted to safe character sets before being used.
[CREDENTIALS_UNSAFE]: The skill includes a dedicated validation script, scripts/config/resolve-config.sh, which performs a security scan on local configuration files. It uses jq to identify and block any files containing keys that resemble API tokens, secrets, or other credentials, ensuring that sensitive information is not accidentally processed or exposed.
[INDIRECT_PROMPT_INJECTION]: The instructions and reference documentation explicitly categorize all data received from external sources (such as GitHub issue titles, comments, and project metadata) as untrusted content. It mandates sanitization and prohibits the agent from interpreting this data as instructions, effectively mitigating indirect prompt injection risks.
[DYNAMIC_EXECUTION]: Configuration resolution is performed dynamically using the jq utility. The process is protected by pre-execution checks that verify input files are regular files and not symbolic links, preventing potential file-based attacks. The Python-based run store reader also employs Unix locking and ownership checks to ensure data integrity during read operations.

woostack-plan