The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_tests.py programmatically executes shell commands to run various testing frameworks.
Evidence: Uses subprocess.run to call pytest, npx jest, and npx vitest based on the detected project environment.
Evidence: Appends user-provided target strings from the input JSON directly into the command list.
[DYNAMIC_EXECUTION]: Several scripts utilize dynamic module loading and file system operations based on runtime identifiers.
Evidence: scripts/skill_learn.py employs importlib.util.spec_from_file_location to load the core/skill_persistence.py module from a path calculated relative to the script location.
Evidence: scripts/skill_learn.py performs file system exports and directory deletions (shutil.rmtree) using a skill_id provided via JSON input.
Evidence: scripts/loop_entry.py modifies sys.path to include parent directories for importing core logic components.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an orchestrator that ingests code implementations and test outputs, creating a vulnerability surface where third-party data could influence agent behavior.
Ingestion points: scripts/run_tests.py captures and parses stdout/stderr from external test processes; SKILL.md describes the flow of analyzing user-provided requirements and technology contexts.
Boundary markers: The skill does not explicitly define delimiters for separating tool outputs or code content from its internal instructional logic.
Capability inventory: The skill has the ability to execute commands, write to the file system, and trigger various MCP tools (mcp__pal__*, mcp__rube__*).
Sanitization: While the scripts perform JSON parsing of inputs, there is no explicit sanitization of the content captured from test runner outputs before it is returned to the agent context.

sc-implement