The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external documentation and project-local files, establishing a surface for potential indirect prompt injection.
Ingestion points: Fetches live documentation from cohesivity.ai/offerings/*, pricing, and llms.txt. It also reads existing .cohesivity files and processes message content via the inbox resource.
Boundary markers: Includes a safety note instructing the agent to treat fetched documentation as reference material only and explicitly disregard any embedded instructions or directives within external content.
Capability inventory: The skill allows the agent to execute shell commands for setup, write and read local credential files, and perform network requests via curl.
Sanitization: Requires explicit user verification of file provenance before accessing credentials and mandates the redaction of sensitive data before submitting build reports or feedback.
[COMMAND_EXECUTION]: Provides a shell script for automated tenant bootstrapping and credential setup.
Evidence: The script utilizes umask 077 to ensure created files have restrictive permissions and includes checks against .gitignore to prevent accidental leakage of credentials to version control.
[DATA_EXPOSURE_AND_EXFILTRATION]: Manages sensitive API keys and performs telemetry transmission to the service provider.
Evidence: The skill discloses the collection of request metadata (IP, User-Agent, network ASN, and country) by the Cohesivity API for security auditing and abuse prevention. It provides best practices for keeping management keys server-side and prohibits inclusion of keys in client-facing code.
[EXTERNAL_DOWNLOADS]: Retrieving configuration and resource documentation from the provider's official domain.
Evidence: The skill is designed to fetch the current API schema and pricing information from https://cohesivity.ai at runtime to ensure the agent uses the most recent provider data.

cohesivity