The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external, untrusted sources such as Gmail, Slack, and webhooks, creating a surface for indirect prompt injection attacks.
Ingestion points: Data enters the agent context via tool outputs from external apps (e.g., email bodies, chat messages) and webhook triggers.
Boundary markers: The skill implements strong boundary instructions in 'Hard Rule 7', explicitly telling the agent to treat outsider-authored content as untrusted data and not as instructions. Examples also include specific prompt wrappers to ignore embedded directives.
Capability inventory: The skill can perform diverse actions through FLOWRA_MULTI_EXECUTE_TOOL, FLOWRA_HTTP_REQUEST, and management tools for workflows and agents.
Sanitization: The instructions require the agent to extract only structured fields and use mandatory human approval gates (human_wait or humanApproval) before executing sensitive actions like sending messages, deleting data, or making payments.
[DYNAMIC_EXECUTION]: The skill supports the creation of custom logic using JavaScript for platform-hosted tools.
Evidence: In references/registry.md, the FLOWRA_CREATE_OR_UPDATE_TOOL tool allows users to define a script consisting of an async function run(state). These scripts are executed within the Flowra platform's hosted sandbox environment.
[COMMAND_EXECUTION]: The skill provides developer instructions for building and linking a CLI tool locally.
Evidence: references/cli.md contains shell commands for installing the CLI from a source directory using pnpm and npm link.
[EXTERNAL_DOWNLOADS]: The skill utilizes vendor-managed packages and repositories for its operation.
Evidence: It suggests installing the @flowra/sdk Node.js package and uses npx skills add flowradev/sdk for skill management. It also communicates with vendor domains including flowra.dev and mcp.flowra.dev.
[PERSISTENCE]: The CLI component maintains local authentication state.
Evidence: According to references/cli.md, the flowra login command persists project API keys and configuration in ~/.flowra/config.json with restricted file permissions.

flowra