The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data which could potentially be manipulated to influence agent behavior.
Ingestion points: The workflow fetches OpenAPI specifications from remote URLs or local files (bun run api:sync) and reads project context maps generated by the project-discovery skill.
Boundary markers: The skill defines a strict boundary in Phase 0-2 (Analysis and Planning) where no writes to the codebase are allowed. Transition to Phase 3 (Execution) is gated by the creation of a plan file (.context/reports/test-framework-adaptation-plan.md) and a requirement for explicit user approval.
Capability inventory: The agent has the capability to write files, execute shell commands (via bun and sh), perform network requests (via curl), and manage GitHub repository secrets.
Sanitization: While the workflow is highly structured, there is no explicit mention of sanitizing the content of external OpenAPI specs or context maps before they are used to generate TypeScript facades or project configurations.
[DATA_EXFILTRATION]: The skill manages the transmission of sensitive environment variables to remote repositories.
Evidence: In Phase 7.2, the skill provides instructions for automatically pushing CI secrets from a local .env file to GitHub repository secrets using the command bunx varlock run -- sh -c 'printf %s "$NAME" | gh secret set NAME'.
Security Control: This capability is documented as an opt-in feature that requires the agent to ask the user first. The instructions also specify using shell pipes (|) to ensure secret values do not appear in the LLM's transcript or command-line history.
[COMMAND_EXECUTION]: The workflow relies on executing several high-privilege shell commands.
Evidence: The skill uses bun run api:sync to fetch API definitions, gh secret set to update repository configuration, and curl to verify authentication tokens against live endpoints during the verification phase (Phase 5.3).
[DYNAMIC_EXECUTION]: The skill generates and adapts code at runtime based on external definitions.
Evidence: Phase 4 involves generating and updating TypeScript type facades (api/openapi-types.ts) based on the fetched OpenAPI source, and Phase 6 involves scaffolding new domain entity components.
[EXTERNAL_DOWNLOADS]: The skill is configured to download resources from external URLs.
Evidence: The api:sync functionality downloads OpenAPI specifications from user-provided URLs. The workflow also references documentation and tools from the upex-galaxy vendor infrastructure (e.g., dojo.upexgalaxy.com).

test-framework-adaptation