The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local Python scripts and modules to perform quality checks and manage state. Evidence includes calls to python scripts/quality_gates/four_us_score.py, python scripts/quality_gates/banned_word_check.py, and the scripts.quality_gates.eco_gate module. These tools are part of the skill's infrastructure and are used to enforce quality standards.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted content which presents a surface for indirect prompt injection.
Ingestion points: Processes scraped pages, competitor copy, ad examples, and search results in SKILL.md.
Boundary markers: Explicitly instructs the agent to treat these external sources as untrusted source material that is outranked by internal repository instructions.
Capability inventory: Has the ability to execute shell commands (via Python) and write files (objective.yaml, state.json) to the local workspace.
Sanitization: Uses independent quality gates and requires quantitative claims to be verified by a "collector" tool rather than raw scraped data.

kai-goal