The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-supplied documentation, such as PRDs and implementation backlogs. These documents serve as an attack surface where adversarial instructions could potentially be embedded to influence the agent's logic during the specification-writing process.\n
Ingestion points: 18_implementation_backlog.md, handoff/STATE.md, and various markdown artifacts within the project's app-spec/ directory.\n
Boundary markers: The framework defines strict sequential structures but does not explicitly implement delimiters or instruction-ignore blocks for user-provided data.\n
Capability inventory: The skill utilizes scripts for automated file system management and project state monitoring via git status.\n
Sanitization: Content is parsed using regular expressions for structured data extraction, though it lacks explicit sanitization for prompt injection within the prose text.\n- [COMMAND_EXECUTION]: The script scripts/build_sequence_view.py executes git status to determine the current progress of the project. The implementation uses a static argument list within a subprocess.run call, which effectively prevents command injection as no user-provided strings are interpreted by the shell.\n- [EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to use npx ctx7@latest to ground architectural decisions in official documentation. This operation retrieves tools from the npm registry, a well-known service for developer packages. Additionally, it references established Flutter libraries for asset automation.

specification-writing