The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is configured to execute local shell scripts and Python scripts as part of its setup and operational flow.
Evidence: bash <plugin-root>/scripts/setup.sh is invoked during the prerequisite phase.
Evidence: Multiple instances of uv run <literature-review-dir>/scripts/X.py are used to interface with research databases and process data.
[DYNAMIC_EXECUTION]: The skill assembles and executes shell commands that incorporate dynamic environment variables and user-defined paths.
Evidence: Command strings are constructed using variables like $WS and ${slug}, such as uv run <literature-review-dir>/scripts/build_corpus.py --output "$WS/corpus.json".
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external scientific databases and web sources, which could potentially contain malicious instructions designed to influence the agent's behavior during synthesis.
Ingestion points: Data is retrieved from OpenAlex, Europe PMC, arXiv, and general web searches via WebSearch and WebFetch tools in the SKILL.md file.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded commands within the retrieved research papers or abstracts.
Capability inventory: The agent has access to Bash (command execution), WebFetch (network operations), and file system tools (Read, Grep, Glob).
Sanitization: There is no mention of sanitization, filtering, or validation of the external text content before it is processed for extraction or synthesis.

systematic-review