The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a metadata file named competitor-trial-monitor_audit_result_v2.json which contains comprehensive "PASS" results for security, stability, and integrity. This is a self-referential injection pattern designed to mislead automated security scanners or human auditors into adopting the skill's own claims of safety rather than performing an independent assessment.
Ingestion points: The skill includes the competitor-trial-monitor_audit_result_v2.json file as part of its distribution, which is processed during security evaluation. It also ingests external trial data from the ClinicalTrials.gov API in scripts/main.py.
Boundary markers: There are no boundary markers or instructions to ignore embedded content within the metadata or the processed API data.
Capability inventory: The skill possesses file-write capabilities (writing to ~/.openclaw/competitor-trial-monitor/) and network-read capabilities (urllib.request.urlopen) in scripts/main.py.
Sanitization: The script scripts/main.py performs basic structural validation by parsing API responses as JSON and extracting specific keys, but the audit metadata file is processed as static content by the environment.
[EXTERNAL_DOWNLOADS]: The script scripts/main.py fetches data from the well-known public service clinicaltrials.gov. This behavior is documented and consistent with the skill's primary purpose of monitoring clinical trials.
[SAFE]: The skill uses standard CLI patterns and stores its persistent data in a dedicated hidden directory (~/.openclaw/competitor-trial-monitor/) which is common practice for developer tools. No privilege escalation, persistence mechanisms, or unauthorized command execution patterns were detected in the source code.

competitor-trial-monitor