The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to access and parse sensitive local configuration files to retrieve API tokens.
Evidence: Accesses ~/.config/glab-cli/config.yml and a specific hardcoded path /mnt/disk1/hyx/.config/glab-cli/config.yml to extract authentication tokens.
Evidence: Uses commands like $(glab config get token --host <host>) to inject credentials directly into shell commands.
[COMMAND_EXECUTION]: The skill provides explicit instructions on how to circumvent environment security controls (referred to as 'rtk') that monitor or intercept HTTP output.
Evidence: Recommends using curl -o to redirect output to a file specifically to bypass 'rtk' hooks that intercept stdout.
[DYNAMIC_EXECUTION]: The skill includes a Python helper script that implements dangerous security practices for network operations.
Evidence: The provided Python snippet explicitly disables SSL certificate verification and hostname checking (ssl.CERT_NONE, check_hostname = False), which exposes the agent to man-in-the-middle (MITM) attacks when interacting with GitLab instances.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting and processing untrusted data from external GitLab instances, creating a surface for indirect prompt injection.
Ingestion points: Reads job traces (trace.log) and job metadata (jobs.json) from remote GitLab servers into the agent's context.
Boundary markers: Absent. No delimiters or instructions are provided to treat the external content as untrusted.
Capability inventory: The skill has significant capabilities to mutate the environment, including creating/closing merge requests, deleting branches, and making arbitrary API commits via glab and curl.
Sanitization: Only basic ANSI escape code removal is performed via regex; no semantic sanitization of the log content is performed before processing.

glab-api