The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands and scripts, including bash docs/demo/mortal-install.sh, python3 docs/audit/check-claims.py, and various subcommands of the of CLI tool (e.g., of doctor, of status, of patch).- [REMOTE_CODE_EXECUTION]: The skill features a release-based update mechanism (bash install.sh --from-release) that downloads and executes code. This risk is managed by requiring user consent before the operation and utilizing SHA256 checksums to verify the integrity of the downloaded artifacts.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external files such as docs/model-catalog.md and docs/efficiency-signal.md. This untrusted data influences the selection of model tiers and harness adapters, which are subsequently passed as arguments to subprocess calls (of patch --model-hints, of pack --model-tier). The instructions mitigate this by mandating that the agent must propose these changes to the user for consent rather than executing them silently.

of