The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install its software using the command curl -fsSL https://pinchtab.com/install.sh | bash in SKILL.md. This pattern executes a script from an external domain directly in the system shell without prior verification, posing a severe risk of arbitrary code execution.\n- [COMMAND_EXECUTION]: The tool includes commands for establishing persistence on the host system, such as pinchtab daemon install mentioned in references/cli-reference.md, which installs the service as a background process.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from the web.\n
Ingestion points: Web content extracted via pinchtab text and pinchtab snap (SKILL.md).\n
Boundary markers: Absent; there are no instructions for the agent to distinguish between its own goals and instructions embedded in the scraped text.\n
Capability inventory: The skill has access to the Bash tool and can perform file uploads (/upload) and downloads (/download) as documented in references/api-reference.md.\n
Sanitization: Absent; the skill does not suggest any filtering or validation of the extracted web content before it enters the agent's context.\n- [COMMAND_EXECUTION]: The skill provides the ability to execute arbitrary JavaScript within the browser context via the pinchtab eval CLI command and the /evaluate HTTP endpoint, which could be exploited to manipulate page data or steal session information.

pinchtab