The Agent Skills Directory

[DATA_EXFILTRATION]: The skill documents the use of crit share, which uploads local files to external servers (Crit deployments). It also uses crit push to transmit local data to GitHub and GitLab repositories.
[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to ingest code review comments from external sources such as local files, GitHub PRs, and GitLab MRs using crit comments and crit pull. This external data could contain malicious instructions intended to manipulate the agent's behavior.
Ingestion points: Data enters the agent's context through crit comments --json, crit pull, and the parsing of review.json files.
Boundary markers: While the instructions utilize structured JSON for parsing, they lack explicit directives to ignore or isolate instructions embedded within the natural language of the comment bodies.
Capability inventory: The skill allows for shell command execution (crit CLI), file writing (creating temporary JSON files for batch comments), and network operations via the CLI tool.
Sanitization: The instructions do not describe any sanitization, filtering, or validation processes for the ingested comment content.
[COMMAND_EXECUTION]: The skill is centered around executing shell commands via the crit CLI. It passes data-derived content, such as comment bodies, as arguments to these commands. The documentation attempts to mitigate shell injection risks by instructing the agent to use single-quotes for comment bodies.

crit-cli