The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill dynamically searches for and sources a shell script (comet-env.sh) from various locations, including hidden directories in the user's home folder. Sourcing a script from a computed path (. "$COMET_ENV") allows for the execution of arbitrary code if a file with that name is present in any of the searched locations.
[COMMAND_EXECUTION]: The skill executes multiple shell commands and scripts ($COMET_STATE, $COMET_HANDOFF, $COMET_GUARD) to manage its workflow and phase transitions.
[DATA_EXFILTRATION]: The skill's search pattern includes sensitive directories such as $HOME/.config and $HOME/.gemini, which potentially exposes configuration data and agent interaction history to the search process.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection via the processing of untrusted project artifacts.
Ingestion points: Reads data from proposal.md, design.md, tasks.md, and specs/*/spec.md in file SKILL.md.
Boundary markers: No delimiters or instructions are provided to the brainstorming tool to ignore embedded instructions in these files.
Capability inventory: The skill can execute shell commands and modify local files.
Sanitization: No input validation or sanitization is performed on the ingested content.

comet-design