The Agent Skills Directory

[PERSISTENCE]: The skill establishes multiple persistence mechanisms to ensure it runs automatically. The setup.sh script installs a macOS launchd agent by creating a property list file at ~/Library/LaunchAgents/com.jianshuo.wjs-tweeting-from-articles.plist, which triggers the daily.sh script every morning at 09:00. Additionally, the SKILL.md file provides instructions and commands to append a cron job to the user's crontab for periodic queue processing.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it incorporates content from external article files directly into a large language model prompt.
Ingestion points: In daily.sh, the skill reads the content of article.md from directories within ~/code/wechat-publish/articles/ or a local mirror directory.
Boundary markers: The article content is interpolated into a heredoc prompt for the claude CLI without any delimiters, escape characters, or instructions to ignore embedded commands.
Capability inventory: The skill has the capability to perform network operations (via the xurl tool to post to X/Twitter) and write data to the local filesystem (via jq appending to history.jsonl).
Sanitization: There is no evidence of sanitization, validation, or filtering of the article content before it is processed by the model.
[COMMAND_EXECUTION]: The skill relies on the execution of several shell scripts and system binaries to perform its tasks. It executes scripts/pick-next-article.sh, daily.sh, and scripts/mirror-articles.py. It also makes frequent use of the xurl utility for network requests, jq for JSON processing, and the claude CLI for text generation.

wjs-tweeting-from-articles