The Agent Skills Directory

[DYNAMIC_EXECUTION]: The execution script scripts/run.py dynamically computes the repository root by searching upwards for an AGENTS.md file to modify the Python system path. This allows the skill to import local modules from the tooling directory. While this facilitates modularity in a repository environment, modifying the module search path at runtime using computed locations is a monitored dynamic loading pattern.
[INDIRECT_PROMPT_INJECTION]: The skill processes research paper records (titles and abstracts) provided in external workspace files, which represents an indirect injection surface.
Ingestion points: Data is ingested in scripts/run.py via the load_candidate_records function which reads files from the workspace.
Boundary markers: The skill does not employ explicit boundary markers or "ignore instructions" delimiters when processing the paper content.
Capability inventory: The skill possesses file-writing capabilities, specifically for creating the papers/screening_log.csv file.
Sanitization: Input text is processed through a tokenization function _match_tokens in scripts/run.py which uses a regular expression ([a-z0-9]+) to filter content, reducing the likelihood of embedded instructions being misinterpreted, though the capability for file-writing remains accessible.

screening-manager