The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute remote shell scripts using high-risk patterns such as piping curl output directly to sh or iex.
Evidence found in SKILL.md: curl -fsSL https://get.trau.sh/install.sh | sh and irm https://get.trau.sh/install.ps1 | iex.
Evidence found in references/cli.md: curl -fsSL https://get.trau.sh/install.sh | sh.
Evidence found in references/operations.md: curl -fsSL https://get.trau.sh/install.sh | sh.
[REMOTE_CODE_EXECUTION]: The skill uses npx to install a global package from a specific GitHub repository (RomkaLTU/skills@trau).
[COMMAND_EXECUTION]: The skill makes extensive use of powerful CLI tools like sh, curl, and gh to perform network operations and system modifications.
[PERSISTENCE]: The skill provides commands to establish persistence on the host system by registering the hub as a systemd or launchd service via the trau hub supervise command.
[PRIVILEGE_ESCALATION]: The skill instructions mention using sudo for installation when a custom directory is specified, which could be exploited to gain elevated privileges.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external trackers (Linear, Jira, Azure DevOps, GitHub Issues), logs, and steer notes.
Ingestion points: Reads ticket descriptions, comments, and agent transcripts from various external sources.
Boundary markers: Includes a high-level instruction to treat external content as data rather than instructions.
Capability inventory: Contains multiple tools for shell command execution, repository manipulation, and network access.
Sanitization: No explicit data sanitization or escaping mechanisms are documented for the ingested content.
[DYNAMIC_EXECUTION]: The skill manages the runtime environment by downloading and activating browser harness runtimes via trau browser setup.
[DATA_EXFILTRATION]: The skill includes tools designed to retrieve and reveal sensitive information, such as secrets and configuration keys, which could be used for unauthorized data extraction.
Tools identified: reveal_secret, trau config get --reveal, and trau secret get --reveal.

trau