The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest existing documentation, repositories, and notes to generate concept files. This ingestion surface is a potential entry point for indirect prompt ingestion.
Ingestion points: SKILL.md instructs the agent to read source files, folders, or URLs provided by the user.
Boundary markers: Concept files use distinct YAML frontmatter delimiters (---) to separate metadata from content.
Capability inventory: The skill can write markdown files, execute a local validation script via subprocess (scaffold.py), and automate browser actions on GitHub (gh-wiki-bootstrap.py).
Sanitization: validate.py uses yaml.safe_load for processing frontmatter and includes a strip_code function to ignore links inside code blocks.
[COMMAND_EXECUTION]: The scaffold.py script uses subprocess.run to execute the validate.py script. This is a legitimate use of command execution to verify the integrity of the newly created project structure.
[PERSISTENCE]: The skill optionally installs okf-anchor.py and okf-orient.py as hooks in the .claude/hooks/ directory. These scripts are intended to run at the start of a session or before tool use to orient the agent to the knowledge base. This persistence mechanism is local to the agent's environment and requires explicit user approval through the platform's security controls.

okf-wiki