The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a feature called build_context_bundle (documented in references/commands.md and references/usage-examples.md) which reads and提炼 (refines) content from user-specified directories containing text, documentation, and images. This content is then injected into the agent's context as a 'Requirement Snapshot' to guide the Research and Planning phases.
Ingestion points: Untrusted data enters the agent context via the build_context_bundle command which processes files from arbitrary local directory paths.
Boundary markers: The instructions do not explicitly specify the use of delimiters or protective 'ignore instructions' wrappers when interpolating this external content into Spec files or the active session context.
Capability inventory: The skill possesses filesystem write capabilities (creating and updating Spec/CodeMap files) and the ability to execute provided local scripts.
Sanitization: There is no evidence of sanitization or safety-filtering performed on the content parsed from the external files before it is processed by the LLM.
[COMMAND_EXECUTION]: The skill incorporates two local Python scripts used for automation tasks: scripts/archive_builder.py (for consolidating project artifacts) and scripts/default_prompt_check.py (for verifying and updating project instruction files). While these scripts execute shell commands via the Python interpreter, they are bundled with the skill, perform localized filesystem operations, and do not utilize network access or unsafe dynamic evaluation of user-controlled strings.

sdd-riper-one