The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes data from untrusted external sources to generate its lessons and exercises.
Ingestion points: The skill reads local user code to generate relevant examples ('grounding') as described in modes/lesson.md and fetches primary sources from the web using WebSearch and WebFetch as specified in SKILL.md.
Boundary markers: SKILL.md defines a 'routing rule' context guard that restricts the agent's file access to a single topic folder per session after routing, which helps limit the scope of potential injections.
Capability inventory: The skill is configured with Bash, Write, Edit, and Agent tools (SKILL.md), allowing it to modify the local filesystem and suggest shell commands.
Sanitization: The instructions do not define any specific sanitization or validation routines for data retrieved from the web or user repositories before it is used in prompt interpolation.
[DYNAMIC_EXECUTION]: The skill generates and facilitates the execution of new code based on the learning topics.
Evidence: modes/lesson.md instructs the agent to write runnable code to an exercises/ directory and output the specific command the user should run to execute that code.
[COMMAND_EXECUTION]: The skill executes shell commands for system integration and repository setup.
Evidence: SKILL.md specifies running git init to initialize the learning repository and utilizes platform-specific commands (such as open, xdg-open, or start) to launch generated HTML artifacts in a browser.
[EXTERNAL_DOWNLOADS]: The skill retrieves external data to populate its knowledge base.
Evidence: The skill uses WebSearch and WebFetch to identify and ingest content from primary sources, which are subsequently cataloged in SOURCES.md as noted in SKILL.md and modes/lesson.md.

tutorkit