The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves ingesting untrusted data from the user's local repository, creating a potential vector for indirect prompt injection if the repository contains malicious instructions.
Ingestion points: Phase 1 ("Discovery & Archeology") in SKILL.md instructs the agent to run ls -R and grep on configuration files and directory structures to understand the tech stack.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" directives to prevent the agent from following commands potentially embedded in file names, directory names, or package.json dependencies.
Capability inventory: The skill requires shell access to perform its discovery functions via ls and grep as specified in SKILL.md.
Sanitization: There are no explicit steps provided to sanitize or validate the data retrieved from the file system before it is used to generate C4 diagrams or Architecture Decision Records (ADRs).
[COMMAND_EXECUTION]: The skill explicitly directs the agent to execute shell commands to perform repository discovery.
Evidence: SKILL.md contains instructions to run ls -R and grep -r "dependencies" package.json, as well as scanning for specific infrastructure files like Dockerfile, terraform/, and k8s/.

senior-solution-architect