The Agent Skills Directory

[SAFE]: The skill establishes a set of best practices and audit procedures for GraphQL API design and implementation. It leverages standard industry libraries and patterns without introducing malicious code.
[COMMAND_EXECUTION]: The documentation in references/contract-gates.md and references/client-tanstack.md describes standard project tasks using cargo, bun, and mise. The main instructions in SKILL.md (Step 5) provide robust safety constraints, requiring explicit task authorization and enforcing read-only environments to prevent unauthorized execution or unintended modifications to the repository.
[INDIRECT_PROMPT_INJECTION]: As a review tool, the skill is designed to ingest external data such as GraphQL diffs and API surfaces. It includes a dedicated references/runtime-trust.md module that instructs the agent to treat all repository content, scripts, and web content as untrusted data, effectively establishing boundary markers and sanitization protocols.
Ingestion points: The skill accepts external GraphQL diffs, modules, or API surfaces as primary input (SKILL.md).
Boundary markers: The references/runtime-trust.md file provides explicit instructions to treat all external data as untrusted and prohibits embedded instructions from altering the agent's scope or authority.
Capability inventory: Command execution is restricted to authorized tasks and Reproduces schema output only in temporary state, avoiding local file writes (SKILL.md Step 5).
Sanitization: The skill mandates that secrets discovered in the code must not be reproduced in outputs and that repository content cannot grant additional authority (references/runtime-trust.md).
[DATA_EXPOSURE]: The skill incorporates a specific security policy in references/runtime-trust.md that forbids copying discovered credentials into outputs, logs, or artifacts, directing them instead to authorized security channels.

tailrocks-graphql-review