The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local bash scripts (fetch-model-data.sh, rubric-path.sh) and a Python validator (resolve-route.py) to perform its internal logic. These scripts are part of the skill's own package and are used to manage user configuration files.
[EXTERNAL_DOWNLOADS]: The skill fetches model pricing and performance data from well-known industry sources including Artificial Analysis and the DeepSWE leaderboard. These external references are used neutrally to provide data for the model routing rubric.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data regarding model benchmarks, which presents a potential injection surface.
Ingestion points: Data is fetched from the Artificial Analysis API and the DeepSWE leaderboard via external fetcher scripts.
Boundary markers: The skill treats data as source evidence for a routing interview rather than direct commands.
Capability inventory: The skill has permissions to write to the model-rubric.yml configuration file and execute shell commands for capability reconciliation (e.g., command -v).
Sanitization: A mandatory validation process is implemented through resolve-route.py validate, which verifies the schema and reachability of the draft rubric before it is written to the filesystem.
[SAFE]: The skill implements strong security guidance regarding credentials, instructing users to add API keys to their local environment files (like .zshenv) rather than providing them in the interactive session.

model-rubric