The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages sensitive platform credentials including API keys and bot tokens. It explicitly instructs the agent to use environment variables for secret ingestion and provides warnings against accidental token disclosure through the --show flag.
[EXTERNAL_DOWNLOADS]: The skill instructions facilitate the installation and update of the trylle-cli binary and related skill components from external registries including NPM (@trylle/trylle-cli), PyPI (via uv), and Cargo. These resources are associated with the official vendor infrastructure.
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands through the try CLI and git binary. This includes high-impact lifecycle operations such as repository deletion, pull request merging, and the modification of CI/CD secrets and variables.
[INDIRECT_PROMPT_INJECTION]: The skill has a high-risk attack surface for indirect prompt injection.
Ingestion points: The agent is instructed to read content from external, potentially attacker-controlled sources including pull request bodies, issue comments, and code diffs (SKILL.md, references/workflows.md).
Boundary markers: The instructions recommend using the --json flag for structured parsing and --body-file for large text inputs, which provides some structural separation but does not prevent instruction following from the content.
Capability inventory: The skill possesses capabilities for repository deletion, pull request merging, and modifying repository secrets (references/command-catalog.md).
Sanitization: There is no mention of sanitization or filtering of the content retrieved from external resources before the agent processes it.

trylle-cli