The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bun for environment execution, testing, and CI tasks. It also leverages bunx --bun shadcn to manage UI components. These are standard developer tools and are used within their intended scope.
[DYNAMIC_EXECUTION]: The skill executes a localized vendor script, resolve-package-versions.ts, from the peer tailrocks-tanstack-project-setup repository to manage dependency resolution. This is a functional requirement for maintaining canonical package versions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository files, external API responses, and templates. It mitigates this risk through the runtime-trust.md reference, which explicitly instructs the agent to treat all repository content as untrusted data and prevents embedded instructions from altering the skill's authority or rules.
Ingestion points: SKILL.md (reading templates and ledger rows), references/boundaries-and-data.md (parsing route params, search params, and external API responses).
Boundary markers: references/runtime-trust.md provides clear markers requiring the agent to disregard instructions within untrusted data.
Capability inventory: File system writes (SKILL.md), network operations via bun (references/tooling-and-quality.md), and UI component management (references/shadcn-ui.md).
Sanitization: references/boundaries-and-data.md mandates that all untrusted values be parsed and validated before use.
[DATA_EXFILTRATION]: The skill implements a robust secret-handling policy in references/runtime-trust.md, prohibiting the copying of credentials into logs, prompts, or artifacts. It further restricts secret access to the server environment via references/boundaries-and-data.md.

tailrocks-tanstack-project-remediate