The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project metadata and usage statistics retrieved from the Sentry API, creating a surface for potential indirect prompt injection if an attacker can control Sentry project names, team names, or settings that are later processed by the agent.
Ingestion points: Data is ingested from the Sentry API (sentry.io/api/0) in audit-sentry-drift.sh and check-team-quotas.ts (scripts intended to be created by the user).
Boundary markers: Data is parsed as JSON, but there are no explicit boundary markers or instructions to the agent to treat this external data as potentially untrusted.
Capability inventory: The skill utilizes Write, Edit, and Bash capabilities to manage configuration files and run audit scripts.
Sanitization: Data is processed using standard JSON parsers and specific field access, but there is no explicit sanitization for LLM-specific injection patterns.
[COMMAND_EXECUTION]: The skill provides scripts and CI workflow templates that execute shell commands and interact with network APIs to perform governance audits and enforce policies.
The audit-sentry-drift.sh script uses curl to fetch project data and python3 for processing and validation.
The check-team-quotas.ts script uses fetch to retrieve usage stats and is executed via npx tsx in CI environments.
CI workflows utilize grep and standard shell utilities to block policy violations such as hardcoded DSNs or direct SDK initialization.

sentry-policy-guardrails