The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads external data from repository files to populate training checklists and identify installed modules, which could be exploited if an attacker can modify those files.
Ingestion points: The skill reads from README.md files, docs/handover.md, and potentially an external configuration directory specified by the $POWERSUIT_OVERLAY environment variable.
Boundary markers: No explicit delimiters or "ignore embedded instructions" warnings are used when the agent processes the content of these files.
Capability inventory: The skill has the capability to write to the filesystem (mkdir, cp), execute shell commands (bash), and run Python scripts (python3).
Sanitization: The skill does not appear to sanitize or escape the content read from repository documentation before using it to generate checklists or instructions.
[COMMAND_EXECUTION]: The skill performs several local command executions to manage the handover process and verify module functionality.
Evidence: Use of bash for directory creation and template copying (mkdir -p docs && cp -n ...).
Evidence: Execution of local repository scripts such as python3 .github/actions/gate/run.py and python3 aieo/probe.py.
Evidence: Use of ls to check for the existence of an overlay directory.
[DYNAMIC_EXECUTION]: The skill uses a Python heredoc to dynamically parse a markdown table at runtime.
Evidence: Step 1 uses `python3
"${CLAUDE_PLUGIN_ROOT}/templates/docs/handover.md" <<'EOF'` to execute an embedded script that parses the "Module signals" section.

owner-training