The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local shell commands for searching content, querying the concept database, and performing validation tasks.
Evidence: Execution of grep, pnpm build:sqlite, pnpm query, pnpm format, pnpm validate, and pnpm build:graph in the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes information from external and potentially untrusted sources (such as documentation repositories and live API probes) to create concept nodes, which could introduce malicious instructions into the agent's workspace.
Ingestion points: Reads from testdata/runpod-migrate/v2-openapi.json, the runpod/docs repository, docs.runpod.io, and live API responses.
Boundary markers: Relies on the pnpm validate --strict tool to enforce schema compliance but does not explicitly define prompt boundary markers for the data ingestion phase.
Capability inventory: Shell command execution via grep and pnpm for repository management, querying, and validation.
Sanitization: Uses pnpm format for canonical data styling and a strict validator to ensure that content is public-facing and safe.
[EXTERNAL_DOWNLOADS]: The skill references and retrieves content from official Runpod documentation for verification and citation.
Evidence: Fetches guidelines and documentation from docs.runpod.io.

add-concept