The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied story content which is then passed to sub-agents for analysis. This data flow creates a surface for indirect prompt injection if the story text contains instructions designed to influence or manipulate the agent's behavior.
Ingestion points: Story files (paths) or raw text provided by the user in Phase 1, Step 3.
Boundary markers: The instructions lack specific boundary markers or warnings for the agent to disregard potential instructions embedded within the story text during analysis.
Capability inventory: The skill writes to the project directory, executes local utility scripts, and delegates analysis tasks to other agent skills.
Sanitization: The skill performs basic text normalization and strips frontmatter but does not implement filtering for malicious prompt patterns.
[COMMAND_EXECUTION]: The skill's operational flow involves executing shell commands to run internal Python and Node.js utility scripts.
Evidence: SKILL.md and references/tracking-transaction.md explicitly direct the agent to run scripts/tracking_commit.py, scripts/wordcount_cli.py, and scripts/check-outline-contract.js using the system's runtime environments.
[DYNAMIC_EXECUTION]: Local Python scripts utilize dynamic module loading at runtime for internal code sharing.
Evidence: scripts/tracking_commit.py and scripts/wordcount_cli.py use importlib.util to dynamically load the wordcount_core.py module from the skill's own directory.

story-import