The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied business descriptions and intents to perform diagnostics. This data is interpolated into tool queries and the final report. While the process is highly structured (Phase 1-4), it lacks explicit delimiters to neutralize potentially malicious instructions embedded within user input.
Ingestion points: User descriptions of their business actions and plans in Phase 1.
Boundary markers: The skill uses a structured Phase-based workflow, but does not wrap user input in XML tags or similar boundary markers to prevent the agent from obeying instructions inside the data.
Capability inventory: The skill executes a local Python script and makes outbound network requests to a vendor API via that script.
Sanitization: The accompanying script uses standard argument parsing but does not perform advanced sanitization of the search terms.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a local Python script, scripts/atoms-search.py, to retrieve content from the atoms.jsonl file or the cloud-based knowledge repository.
[EXTERNAL_DOWNLOADS]: The script atoms-search.py is configured to connect to https://api.xyskill.xyz to perform search operations if the local database is insufficient. This represents standard communication with the vendor's backend infrastructure.
[DATA_EXFILTRATION]: The script accesses local configuration files (e.g., ~/.xy/config.json) to resolve paths and transmits the user's extracted search keywords to the vendor's search API. No sensitive data beyond search queries is transmitted.

xy-slow-lane