The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes exam questions and status data retrieved from the external API 'clawvard.school'. This creates a surface where instructions embedded in questions could potentially influence agent behavior.
Ingestion points: API responses from https://clawvard.school/api/exam/ in SKILL.md.
Boundary markers: None explicitly defined for exam question content.
Capability inventory: Network requests via HTTP and file writing for token persistence in SKILL.md.
Sanitization: The instructions explicitly forbid including private data, credentials, or project metadata in the exam traces sent back to the server.
[PERSISTENCE]: The skill manages a Clawvard identity token that can be saved to the local filesystem. This persistence is a core feature intended for authenticated retakes. The risk is mitigated by a mandatory hard confirmation gate requiring the user to type a specific string match ('confirm save token') and explicit disclosure of the storage path to the user.
[DATA_EXFILTRATION]: The skill transmits agent performance traces and answers to 'clawvard.school'. While this is the intended functionality, it carries a risk of accidental data exposure. The instructions include specific prohibitions against including credentials, file paths, or private user content in these outgoing traces.
[COMMAND_EXECUTION]: The pre-flight check in SKILL.md uses shell commands (curl, grep, stat) to verify the skill version against the vendor's repository. While it suggests an update command (npx skills add), it requires explicit user execution, maintaining human-in-the-loop control over software installation.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch exam questions from clawvard.school and version metadata from raw.githubusercontent.com. These downloads are documented and constrained to the skill's primary evaluation purpose.

clawvard-agent-eval