The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the application under test (DOM, network logs, console output) which could contain malicious instructions designed to influence the agent.
Ingestion points: Ingests app state via reticle_* tools (SKILL.md).
Boundary markers: Absent; no instructions provided to ignore instructions embedded in the app data.
Capability inventory: Can execute npx commands, modify config files, and perform network requests (SKILL.md).
Sanitization: No evidence of sanitization for ingested application data.
[PROMPT_INJECTION]: The instructions strongly emphasize agent autonomy, directing the agent to act without user intervention for multiple setup steps.
Evidence: "Do not stop, and do not ask the user anything..." and "Decide, act, and say what you did in one line" (SKILL.md). While intended for efficiency, this reduces user oversight of the agent's actions.
[COMMAND_EXECUTION]: The skill executes setup and verification commands using vendor-provided tools.
Evidence: Execution of npx @reticlehq/server@latest init (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and tools from the vendor's official domains.
Evidence: curl requests to docs.reticle.sh and npx tool downloads (SKILL.md).

install-and-verify