The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple project documents to generate its technical baseline, creating a vulnerability surface where malicious instructions in those documents could influence the agent's behavior.
Ingestion points: The skill reads and greps content from design/gdd/*.md, docs/engine-reference/, and project.yaml in Phase 0b.
Boundary markers: While it uses structural grep patterns (e.g., ^## Detailed Rules) to target specific sections, it lacks explicit delimiters or instructions to ignore embedded natural language directives within the extracted text.
Capability inventory: The skill possesses significant capabilities including Write, Edit, Bash, and the Agent tool for multi-agent orchestration.
Sanitization: There is no evidence of sanitization or filtering applied to the extracted requirements before they are incorporated into the prompt context.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context execution to hydrate the agent's environment with configuration data at load time.
Evidence: !bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys review_mode,automation,workflow,docs.density in SKILL.md (Line 11).
Assessment: This is used to resolve environment-specific settings (e.g., docs.density, workflow) using a project-local script. It does not appear to exfiltrate data or access sensitive user files.
[COMMAND_EXECUTION]: The skill executes local scripts to manage system registries and configuration resolution.
Evidence: Calls to .claude/hooks/yaml-helper.sh and .claude/scripts/registry.sh via the Bash tool.
Assessment: These operations are restricted to specific project-relative paths as defined in the allowed-tools frontmatter, supporting the skill's infrastructure for requirement tracking and configuration management.

create-architecture