The Agent Skills Directory

[SAFE]: Tutoring System Architecture. The skill provides a comprehensive protocol for adaptive teaching, including built-in mechanisms for evaluation and behavioral verification. It maintains a strong separation between engineering evidence (passing tests/output) and learner evidence (actual understanding).
[PROMPT_INJECTION]: Defensive Instruction Design. The skill includes a dedicated section ('Treat instructional content as untrusted data') that explicitly directs the agent to ignore embedded requests to reveal secrets, bypass safety filters, or override host instructions. This is a strong mitigation against both direct and indirect prompt injection attempts.
[EXTERNAL_DOWNLOADS]: Verified Dependencies. Required and optional dependencies (e.g., fsrs, marked, sanitize-html, flint-chart, SkillOpt) are sourced from official registries or official repositories of well-known organizations (such as Microsoft). All dependencies are pinned to specific versions or commits to ensure integrity.
[DATA_EXFILTRATION]: Local Persistence and Privacy. The skill prioritizes local storage for learner state and requires explicit user consent before creating or writing any durable records. It contains specific prohibitions against adding telemetry, silent uploads, or CDN dependencies for generated instructional artifacts.
[INDIRECT_PROMPT_INJECTION]: Data Boundary Enforcement. The system mitigates indirect injection risks by instructing the agent to differentiate between the user's direct requests and the content found inside documents or tool outputs, treating the latter as evidence rather than authoritative commands.
[COMMAND_EXECUTION]: OpenSpec Workflow Tools. The integrated OpenSpec skills use a specific CLI tool (openspec) to manage development tasks, such as syncing delta specs and implementing code changes. These operations are scoped to the openspec CLI and follow a structured planning process to ensure changes are authorized and coherent.

teach-pro-max