The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a bash script 'scripts/cosense_api.sh' that executes 'curl' commands to interact with the Cosense API and 'python3' for JSON manipulation. These operations are restricted to the local environment and the official service domain.- [DATA_EXFILTRATION]: Sensitive credentials like 'COSENSE_SID' are managed as environment variables and are only transmitted to the legitimate 'https://scrapbox.io' domain for authentication. There is no evidence of data being sent to unauthorized third-party servers.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it retrieves content from external Cosense pages (Ingestion: 'get-page' and 'get-page-text' in 'scripts/cosense_api.sh'). It has the capability to modify remote data (Capability: 'safe-import' in 'scripts/cosense_api.sh'). To mitigate this, the skill instructions mandate that the agent must display changes and obtain user approval before proceeding (Boundary: 'Data safety rules' in 'SKILL.md'). No automated sanitization is performed on the ingested content.- [SAFE]: The skill demonstrates a clear and legitimate purpose with appropriate safety guardrails. No obfuscation, persistence mechanisms, or unauthorized privilege escalations were found in the provided files.

cosense