The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content, including pull request code and feedback comments, which constitutes a typical surface for indirect prompt injection.
Ingestion points: Invocation routes in SKILL.md (e.g., review PR #123) and feedback synthesis in references/reporting.md involve reading data from external pull requests.
Boundary markers: The skill documentation (references/aggregation.md) includes an explicit directive that untrusted data encountered during analysis must not override the agent's core instructions, goals, or security policies.
Capability inventory: The skill possesses the ability to execute shell commands (uv run), write files (reports), and delegate publication tasks to the ship skill.
Sanitization: The references/reporting.md file specifies the use of a sanitized report renderer to prevent cross-site scripting (XSS) and other injection attacks within the generated HTML artifacts.
[DYNAMIC_EXECUTION]: The skill employs dynamic loading techniques to integrate with shared reporting functionality located in sibling directories.
Evidence: The tests/test_report.py file uses importlib.util.spec_from_file_location to load a Python module from a path constructed relative to the current file (../../elenctic/scripts/report.py).
[COMMAND_EXECUTION]: The skill's instructions include the execution of shell commands to process review data and generate report files.
Evidence: references/reporting.md contains an instruction for the agent to execute a Python script via the uv tool (uv run --no-project python <installed-elenctic>/scripts/report.py).

ergasterion