The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill directs the agent to download and install the strava-pp-cli tool using npx from the @mvanhorn/printing-press-library package on NPM and go install from the vendor's GitHub repository. These resources are provided by the vendor to enable the skill's training analysis capabilities.
[DATA_EXPOSURE_AND_EXFILTRATION]: The CLI tool includes a --deliver webhook:<url> flag, allowing output—which may contain sensitive heart rate, power, and location data—to be POSTed to any external HTTP endpoint. This documented feature poses a data exfiltration risk if an agent is manipulated into using it maliciously.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the external Strava API. 1. Ingestion points: Activity names, descriptions, and segment details. 2. Boundary markers: Absent; data is ingested directly into the agent context. 3. Capability inventory: Ability to perform network POSTs via webhooks, write local files, and modify cloud activity data. 4. Sanitization: No sanitization of external strings is mentioned in the instructions.

pp-strava