The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash to create directory structures and write initial configuration files. It uses a defensive pattern of passing user-supplied values through environment variables to a Python one-liner to prevent shell injection during config generation.
[COMMAND_EXECUTION]: The workflow involves executing local Python scripts within the ${CLAUDE_PLUGIN_ROOT}/scripts/ directory to manage binding manifests and wiki metadata.
[PROMPT_INJECTION]: The skill processes untrusted grounding material provided by the user (file paths, text, or URLs) in Step 2.5. This content is read into the agent context (up to 50 KB) to inform the charter-framing process, presenting an indirect prompt injection surface.
Ingestion points: Grounding material provided by the user in Step 2.5.
Boundary markers: None specified for the read operations.
Capability inventory: Bash, Read, Glob, WebSearch, Skill.
Sanitization: Implements a 50 KB sample limit for the ingested content.
[EXTERNAL_DOWNLOADS]: Employs the WebSearch tool to gather preliminary domain information used to suggest initial themes for the knowledge base charter.

knowledge-setup