The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands and a Python script using hardcoded absolute paths tied to a specific user environment (/Users/dan/Documents/ws/search-microsims/...). This involves running a search utility and logging timestamps to the filesystem using bash.
[INDIRECT_PROMPT_INJECTION]: The workflow relies on reading and processing external project files, such as docs/course-description.md, docs/learning-graph/learning-graph.json, and docs/glossary.md, to guide the content generation process. These files serve as ingestion points for untrusted data that is subsequently interpolated into prompts for the agent and its sub-agents without mentioned sanitization or specific boundary markers.
Ingestion points: Step 1.3: Read Shared Context (SKILL.md).
Boundary markers: None identified in the instructions for processing external text.
Capability inventory: The skill can spawn multiple sub-agents using the Task tool, write to local chapter files and logs, and execute local Python scripts.
Sanitization: No explicit sanitization or escaping of the ingested project data is described before prompt interpolation.
[DATA_EXFILTRATION]: The skill accesses project-level metadata, learning graphs, and course descriptions to establish shared context. While these operations are functional, they involve the automated reading and processing of local file system data into the agent's context, which could lead to unintended exposure if the agent is directed to send this context to an external endpoint.

chapter-content-generator