The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install software by downloading a shell script from a remote URL and piping it directly to the system shell. This is a critical security risk as the script content is not verified before execution.
Evidence: curl -fsSL https://raw.githubusercontent.com/daggerhashimoto/openclaw-nerve/master/install.sh | bash in SKILL.md.
Evidence: curl -fsSL https://raw.githubusercontent.com/daggerhashimoto/openclaw-nerve/master/install.sh | bash -s -- --branch next in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill requires cloning and installing dependencies from a third-party GitHub repository (daggerhashimoto/openclaw-nerve) that is not identified as a trusted source.
Evidence: git clone https://github.com/daggerhashimoto/openclaw-nerve.git followed by npm install in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface where agent-generated output is parsed for specific markers to trigger UI rendering (charts, kanban boards) and text-to-speech synthesis. Maliciously crafted data processed by the agent could trigger unintended cockpit behavior.
Ingestion points: Agent responses containing {{CHART_START}}, {{KANBAN_START}}, {{TTS_START}}, {{DIFF_START}}, and {{CRON_START}} markers in SKILL.md.
Boundary markers: The skill uses double-curly-brace markers to delimit data, but lacks explicit safety warnings to the agent to ignore embedded instructions within data payloads.
Capability inventory: The cockpit executes rendering logic, performs text-to-speech synthesis, and handles task scheduling (Cron) based on agent output.
Sanitization: No sanitization or schema validation processes are described for the JSON payloads extracted from the markers.

openclaw-nerve-cockpit