The Agent Skills Directory

[PERSISTENCE]: The skill implements a persistence mechanism on macOS systems by installing a launchd agent. The setup.sh script creates and loads com.jianshuo.wjs-promoting-skills.plist into ~/Library/LaunchAgents/, ensuring that the daily.sh marketing workflow executes automatically every day at 4:00 AM.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its requirement to fetch and process data from external, untrusted sources for market research.
Ingestion points: Marketplace data is gathered from sites like ClawHub, Reddit, and Hacker News via WebFetch and WebSearch in the research-marketplaces.sh script and stored in state/research.md.
Boundary markers: The prompts in prompts/daily-post.md and prompts/community-drafts.md incorporate the research.md content directly into the LLM context without using delimiters or instructions to ignore potential injections within that data.
Capability inventory: The environment allows Write access to the local filesystem and controlled Bash execution (specifically git, jq, stat, and date).
Sanitization: There is no evidence of filtering or validation applied to the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill performs automated command execution to manage its state and orchestration. It invokes the claude CLI in a headless, non-interactive mode and uses shell scripts to execute specific system commands, including git for repository metadata, jq for JSON manipulation, and xurl for social media interactions.

wjs-promoting-skills