The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted tweet content by interpolating it directly into LLM prompts in scripts/src/extractor.ts and scripts/src/icp-filter.ts for role and company identification.
Ingestion points: Scraped tweet text and author bio fetched from X/Twitter via the Apify actor in scripts/src/scraper.ts.
Boundary markers: Tweet content is wrapped in quotes (e.g., Tweet Text: "${tweetText}"), but there are no explicit system instructions provided to the model to ignore instructions potentially embedded within the tweet text.
Capability inventory: The extracted data is written to local JSON files (radar-jobs.json and openclaw-icp-jobs.json) using fs.writeFileSync. The skill does not execute the scraped data as code or shell commands.
Sanitization: No sanitization or filtering is performed on the tweet text before interpolation into the prompt.
[COMMAND_EXECUTION]: The skill includes scripts/run_pipeline.sh and instructions in SKILL.md that execute npm install and npx ts-node. These commands are used to install dependencies and run the skill's local TypeScript project.
[EXTERNAL_DOWNLOADS]: The skill relies on several external components:
Node.js packages are downloaded from the npm registry as specified in scripts/package.json.
A specific remote Apify actor ghSpYIW3L1RvT57NT is invoked in scripts/src/scraper.ts and scripts/src/debug.ts to perform the scraping.
The SKILL.md file references third-party plugins TweetClaw and Hermes Tweet from GitHub as optional components.

twitter-GTM-find-Skill