The Agent Skills Directory

[PERSISTENCE]: The skill provides an option to install a scheduled GitHub Action workflow (.github/workflows/tend-sweep.yml). This mechanism is strictly opt-in, requires the user to select a cadence, and mandates a manual commit by the user, ensuring full transparency and control over the scheduled execution.
[DYNAMIC_EXECUTION]: The skill runs a bundled Node.js script (scripts/validate-config.mjs) to perform structural validation of the generated configuration file. This script executes locally on the host to ensure the output matches the required schema.
[INDIRECT_PROMPT_INJECTION]: The skill reads project-specific files such as package.json, lockfiles, and configuration scripts (svelte.config.js, vite.config.ts) to determine the repository's stack and capabilities. This creates a surface where repository content could influence agent behavior.
Ingestion points: Project configuration files including package.json, pnpm-lock.yaml, svelte.config.js, vite.config.ts, biome.json, and .env.example.
Boundary markers: Instructions specify that the subagent should report findings without acting, and the skill includes explicit 'NEVER' rules to prevent unintended actions based on inferred data.
Capability inventory: The skill has access to Bash (for node, pnpm, npx, and git) and Write tools to update the repository configuration.
Sanitization: The skill relies on instructional constraints and internal 'NEVER' rules to mitigate risks from ingested data, rather than programmatic content filtering.

tend-onboard