The Agent Skills Directory

[PROMPT_INJECTION]: The skill reads source code and injects it into the prompts of sub-agents as context, creating an indirect prompt injection surface. \n
Ingestion points: SKILL.md (Section 1.5) reads target module files using cat api/src/modules/[module]/**/*.ts and cat front/src/modules/[module]/**/*.{vue,ts}.\n
Boundary markers: Absent; the orchestrator does not use delimiters or provide explicit instructions to sub-agents to ignore instructions embedded in the source code.\n
Capability inventory: references/refactor-backend-agent.md and references/refactor-frontend-agent.md define agents with Edit, Write, Task, and Bash tools, allowing them to modify the local environment.\n
Sanitization: No sanitization or filtering is performed on the ingested code before it is passed to sub-agents.\n- [COMMAND_EXECUTION]: The skill executes shell commands to verify the local environment and run build processes.\n
Reads ~/.claude/settings.json to check for the CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS flag, which involves accessing configuration files outside the project scope.\n
Executes project-defined scripts such as bun test, bun run build, and bunx vue-tsc which depend on the local project's package.json configuration.

peach-team-refactor