The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill reads sensitive authentication tokens and client secrets from a designated credentials directory to facilitate API access.
Evidence: Accesses credentials/github_token using cat.
Evidence: Parses credentials/calendar_tokens.json and credentials/client_secret.json using jq to extract access tokens and OAuth credentials.
[DATA_EXFILTRATION]: The skill performs network operations that involve sending local credentials to external service providers.
Evidence: Transmits authentication tokens to Google's API endpoints (googleapis.com and oauth2.googleapis.com) to manage calendar data.
Evidence: Transmits a personal access token to GitHub's API (api.github.com) to retrieve repository and commit information.
[COMMAND_EXECUTION]: The skill incorporates shell-based automation to interact with the system environment and remote services.
Evidence: Employs curl for making network requests to third-party APIs.
Evidence: Uses jq for processing JSON-formatted configuration and API response data.
Evidence: Uses date and cat for timestamp management and local file reading.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data from external sources.
Ingestion points: The skill fetches commit messages from GitHub repositories and event titles/descriptions from Google Calendar via API calls.
Boundary markers: Absent. There are no delimiters or specific instructions to ensure the agent ignores malicious commands embedded in the retrieved content.
Capability inventory: The agent has the ability to perform network requests (curl) and read local files (cat).
Sanitization: Absent. The skill summarizes and formats external data directly into a final report without implementing filtering or escaping mechanisms.

meeting-prep