The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a vulnerability surface for indirect prompt injection within the data subject deletion functions.
Ingestion points: The userId variable in the handleDeletionRequest function (SKILL.md) and the deleteUserData function (references/gdpr-compliance.md) is sourced from user input.
Boundary markers: There are no delimiters or warnings to ignore embedded instructions within the user-supplied data.
Capability inventory: The skill uses fetch and curl to perform network operations against the Sentry API, including searching for and deleting issues.
Sanitization: No validation or sanitization is performed on the userId before it is interpolated into the Sentry query string (?query=user.id:${userId}), which could allow an attacker to craft a query that matches and deletes unintended data.
[DATA_EXFILTRATION]: The skill uses network operations to interact with the Sentry API.
The instructions involve using curl and fetch to send requests to sentry.io for issue deletion and audit log retrieval.
These operations use the SENTRY_AUTH_TOKEN, which is an administrative credential.
As sentry.io is a well-known service, these operations are documented as part of the intended functionality and do not escalate the verdict.

sentry-data-handling