The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a local shell script (orchestrate.sh) to perform provider detection and orchestrate multi-agent workflows.
[DATA_EXFILTRATION]: The skill is designed to transmit user-provided prompts and task details to several external third-party services, including OpenAI (via Codex), Google (via Antigravity), OpenRouter, and OrcaRouter.
[METADATA_POISONING]: The documentation references non-existent or futuristic model versions like gpt-5.6-sol and gpt-5.6-terra, which is deceptive and could lead users to overstate the skill's actual capabilities.
[INDIRECT_PROMPT_INJECTION]: The multi-stage orchestration (Double Diamond, Grapple) creates an attack surface where malicious output from one external model could influence subsequent stages of the workflow. 1. Ingestion points: Data enters the context via the outputs of external LLMs during the probe, grasp, tangle, and grapple phases. 2. Boundary markers: The provided instructions do not define delimiters or isolation protocols for data returned from external providers. 3. Capability inventory: The skill has the ability to execute shell commands via the orchestrator script and write files to the local results/ directory. 4. Sanitization: There is no evidence of sanitization or validation of external model outputs before they are used in subsequent orchestration steps.

skill-parallel-agents