The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze and act upon project configuration files such as package.json, bunfig.toml, and tsconfig.json. These files, which can be modified by external contributors in a monorepo or open-source context, serve as ingestion points for untrusted data. Since the agent is also empowered to execute commands like bun test and bun build --compile based on this configuration, it creates a potential surface for instructions hidden in configuration comments or metadata to influence agent behavior.
Ingestion points: The agent workflow in SKILL.md directs the agent to inspect package.json, bunfig.toml, and tsconfig.json. These are also provided as templates in references/templates/bun-project-setup.md.
Boundary markers: There are no instructions for the agent to use delimiters or ignore embedded instructions when reading these configuration files.
Capability inventory: The skill provides instructions for bun run, bun test (in references/bunfig-test.md), and binary compilation via bun build --compile (in references/build-compile.md).
Sanitization: The skill does not contain instructions for validating or sanitizing the content of the configuration files before the agent processes them.

ts-runtime-bun