The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md instructions and setup-proxy.sh script execute various shell commands to inspect local environments and write configurations.
Evidence: Multiple uses of cat, jq, grep, awk, and curl to read/modify configuration files such as ~/.claude/settings.json, ~/.codebuddy/models.json, etc.
Evidence: The script spawns curl processes to perform health probes against user-provided proxy URLs, including the transmission of the USER_KEY in headers.
[PRIVILEGE_ESCALATION]: The skill performs file system modifications and permission changes necessary for specific agent requirements.
Evidence: setup-proxy.sh executes chmod 700 on directories and chmod 600 on credentials files (specifically for the dsh agent), which is a localized privilege modification.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from agent configuration files and panel APIs which could technically contain malicious instructions, though the impact is limited by the skill's specific execution context.
Ingestion points: Reads local JSON/TOML/YAML configuration files and fetches data from remote Panel APIs (/api/v1/meta/team/list, etc.).
Boundary markers: None identified in the script logic; content is parsed directly via jq or grep.
Capability inventory: Capability to write to sensitive configuration files and execute network requests via curl.
Sanitization: Uses jq for JSON construction which provides some structural safety, but raw values from user input or API responses are interpolated into configuration templates.
[DATA_EXFILTRATION]: While the skill primarily transmits data to the user-specified proxy, it reads sensitive information (API keys, instance IDs) from local configuration files during the scanning phase.
Evidence: The scanning logic in setup-proxy.sh extracts ANTHROPIC_AUTH_TOKEN, apiKey, and experimental_bearer_token from local files to display to the user or reuse in probes.

setup-proxy