The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the ! syntax at the beginning of the file to execute a shell script (yaml-helper.sh) when the skill is loaded. This allows for dynamic context generation from the output of bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config --keys automation before the agent processes the markdown body.
[COMMAND_EXECUTION]: The skill is configured to use the Bash tool to run a local configuration script. The frontmatter restricts this tool to a specific path and subcommand: bash "*/.claude/skills/onboard/../../hooks/yaml-helper.sh" resolve_config *.
[INDIRECT_PROMPT_INJECTION]: The skill scans multiple directories and reads files such as CLAUDE.md, agent definitions, and various project documentation (design, narrative, QA) to generate an onboarding report. If these project files are compromised, they could contain instructions that influence the agent's behavior during report generation.
Ingestion points: Reads from CLAUDE.md, .claude/agents/, design/, tests/, and production/ directories.
Boundary markers: None identified in the prompt templates to distinguish between instructions and ingested data.
Capability inventory: Uses Read (file access), Write (to save the report), and Bash (limited to the config helper).
Sanitization: No explicit sanitization or validation of the content read from the project files was detected.

onboard