The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/spawn-course.sh contains a helper function jqget that utilizes python3 -c with an eval() call to parse JSON output from the herdr tool. This is used to extract fields like tab_id and pane_id from the tool's response.
Evidence: jqget() { python3 -c "import json,sys; d=json.load(sys.stdin); print(eval(sys.argv[1]))" "$1"; } in scripts/spawn-course.sh.
Context: While eval() is generally considered risky, in this implementation, it is used with hardcoded string literals within the script to traverse a dictionary object, significantly limiting any potential for arbitrary code execution.
[COMMAND_EXECUTION]: The skill relies on several shell scripts (scaffold-course.sh, spawn-course.sh, verify-course-links.sh) that perform filesystem operations and execute the herdr CLI tool to manage agent sessions.
Evidence: Use of mkdir, ln -s, cp, mv, and herdr agent start/prompt/list across the provided scripts.
Context: These operations are core to the skill's functionality as an orchestrator and are confined to the academy workspace.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the filesystem, such as PLAN.md, UNDERSTANDING-MAP.md, and RESOURCES.md, to determine the state of learning courses and route sessions.
Ingestion points: SKILL.md describes surveying course files: read PLAN.md (frontier), UNDERSTANDING-MAP.md (status summary), and the newest session-log/*.md (date).
Boundary markers: No specific boundary markers or 'ignore' instructions are used when reading these files in the provided scripts.
Capability inventory: The skill can execute shell commands and interact with other agents through herdr.
Sanitization: The scripts do not perform explicit sanitization of the content read from these files before using them in summaries or orchestration logic.

academy