The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded database connection string containing a password (SUPERSECRETPW) is present in references/scrubbing.md. Although explicitly labeled as a test fixture for measuring scrubber effectiveness, hardcoding credentials in skill files is a discouraged practice.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download the Sentry CLI installation script from https://cli.sentry.dev/install. This is an official domain for Sentry, which is a well-known service.
[REMOTE_CODE_EXECUTION]: The installation instructions for the Sentry CLI involve piping a remote script directly to a shell (curl ... | bash) in references/setup.md. While the source is a well-known service, this pattern is a high-capability execution method.
[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to ingest and process data from external Sentry APIs and CLI outputs, creating a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through sentry project view, sentry api, and sentry issue list commands, as well as the Sentry MCP server.
Boundary markers: No explicit markers or instructions are used to distinguish external data from system instructions or to warn the agent about embedded content in error events.
Capability inventory: The skill utilizes sentry-cli for project and release operations and python3 -c for JSON processing.
Sanitization: The skill employs standard JSON parsing for API responses but lacks specific sanitization logic for the textual content of error messages or project metadata before they are interpreted by the agent.

error-tracking