The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts to interface with various model runners. These scripts are part of the platform's skill infrastructure and are used to orchestrate the multi-model consensus process.
Evidence: SKILL.md and references/runner-invocations.md document shell commands like python3 .agents/skills/codex-runner/scripts/run_codex.py with parameter passing for timeouts, schemas, and output files.
[PROMPT_INJECTION]: The skill passes raw user-provided prompts verbatim to multiple sub-agents. This creates a surface for indirect prompt injection where a malicious task could attempt to influence the models' behavior. However, the orchestrator enforces --restrict-tools and validates output against fixed JSON schemas, which restricts the potential impact to the content of the synthesis rather than system integrity.
Evidence: SKILL.md Phase 1 instructions specify sending the <raw prompt verbatim> to all available seats.
[EXTERNAL_DOWNLOADS]: The skill references and interacts with several well-known AI services and models (Claude, Codex, Gemini, Kimi) through authorized runner scripts and sub-agent tools.
Evidence: SKILL.md defines a seat table including Opus 4.8, Sonnet 4.6, Codex, Gemini, and Kimi.

models-round-table