The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest and process arbitrary web content from external URLs. This creates a significant surface for indirect prompt injection attacks where malicious instructions could be embedded in scraped pages. The skill acknowledges this risk in references/scraping-policy.md and provides sanitization utilities in scripts/lib/text.mjs (e.g., stripTags, cleanForAgent) to mitigate it.
[DYNAMIC_EXECUTION]: scripts/corpus-run.mjs allows the agent or user to execute local JavaScript analysis scripts against the scraped corpus using import(). Although the safeScriptPath function in scripts/lib/bridge.mjs contains comments regarding path restrictions, the implementation currently only enforces file existence and extension checks, potentially allowing the execution of any local .js or .mjs file accessible to the process.
[COMMAND_EXECUTION]: The skill uses child_process.spawnSync in scripts/lib/client.mjs to execute system commands like lsof and ps for managing local Chrome processes. It also invokes Node.js scripts from a sibling skill (octocode-chrome-devtools) to perform browser-based scraping via the Chrome DevTools Protocol (CDP).
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill interacts with external services, specifically api.scrapingant.com, for hosted scraping tasks. To prevent accidental leakage of sensitive information, scripts/lib/bridge.mjs implements robust redaction logic that identifies and masks authorization headers, cookies, and common secret patterns in URLs and request bodies before they are stored in the local corpus.

octocode-scraping