The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface. 1. Ingestion points: Untrusted data enters via the --input and --packet arguments in scripts/ollama-worker.sh. 2. Boundary markers: While the prompt uses labeled sections, it does not explicitly instruct the model to ignore instructions embedded within the data. 3. Capability inventory: The orchestration scripts invoke shell commands (ollama, curl) and perform filesystem writes. 4. Sanitization: The skill lacks sanitization of external content before interpolation into prompts.
[COMMAND_EXECUTION]: The skill executes the ollama CLI and curl through shell scripts to manage and communicate with local models. Evidence: scripts/ollama-health.sh and scripts/ollama-worker.sh.
[DYNAMIC_EXECUTION]: The skill uses node -e in scripts/ollama-worker.sh to dynamically resolve paths and construct JSON payloads. This constitutes runtime script generation and execution. Evidence: scripts/ollama-worker.sh Node.js calls.

octocode-subagent