The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted sources including Reddit subreddits, Hacker News, and various web pages via WebFetch and WebSearch. This content is summarized and written directly to local markdown files (ads/references/<platform>-changelog-30d.md and ads/references/<platform>-audit.md).
Ingestion points: Step 3 fetches JSON data from reddit.com and hn.algolia.com via curl. It also fetches webpage content via WebFetch and WebSearch tools.
Boundary markers: No explicit markers or "ignore embedded instructions" warnings are used when interpolating this untrusted data into the generated markdown reports.
Capability inventory: The skill possesses the Bash tool for command execution and Write/Edit tools for modifying the local filesystem.
Sanitization: There is no evidence of filtering or sanitization of the fetched external content (e.g., stripping markdown injection or hidden instructions) before it is committed to the project files.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute local Python scripts and embedded Python code blocks to manage configuration and process data.
Evidence: Step 2 executes a local script via python3 scripts/ads_sources.py. Step 4 uses python3 -c to run a multi-line script that imports from a local lib directory to rank and deduplicate the fetched results.

ads-update