The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically identifies the installation path for the optional archify tool by searching through a set of common agent skill directories (e.g., ~/.claude/skills, ~/.agents/skills). Once a location is identified, it executes the tool's binary using a computed path (node <archify-skill-path>/bin/archify.mjs). This execution pattern introduces a risk where a malicious actor could place a deceptive script in one of the scanned directories to achieve code execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from various sources, which presents an attack surface for indirect prompt injection.
Ingestion points: Data is read from files in docs/architecture/, .specs/ files, and metadata such as GitHub Pull Request and Issue bodies as part of the gap analysis process.
Boundary markers: The skill includes a specific guardrail in references/gap-audit-handoff.md that instructs the agent to treat untrusted inputs like Issue/PR bodies as data rather than instructions.
Capability inventory: The skill has the capability to write files, create directories, execute external scripts via node, and delegate tasks to other functional skills such as /gap-analysis.
Sanitization: The skill relies on high-level prompt instructions to mitigate injection risks; it does not implement explicit code-level sanitization or schema validation for the ingested content.
[COMMAND_EXECUTION]: The skill utilizes a shell script, scripts/architecture-doctor.sh, which executes various system commands including find, sed, grep, and wc to perform diagnostic environment checks and inventory the architecture documentation.
[EXTERNAL_DOWNLOADS]: The skill references external tools and repositories. It describes a fallback mechanism that uses npx to fetch the @mermaid-js/mermaid-cli package if the local binary is unavailable, and it references a third-party GitHub repository for the optional archify tool.

architecture