The Agent Skills Directory

[COMMAND_EXECUTION]: Several scripts in the ars/scripts/ directory use subprocess.run() to execute system utilities. For example, scripts/check_v3_6_8_pattern_protection.py uses git log and git show to verify the integrity of agent prompts. Additionally, the scripts/run_codex_audit.sh wrapper dispatches the codex CLI for cross-model auditing of deliverables.
[EXTERNAL_DOWNLOADS]: The suite makes legitimate network requests to trusted academic databases, including Semantic Scholar (api.semanticscholar.org), OpenAlex (api.openalex.org), and Crossref (api.crossref.org). These requests are used for 'Tier 0' programmatic reference verification and the contaminated-source triangulation feature.
[PROMPT_INJECTION]: The skill's agent prompts include role-playing instructions and defensive markers like 'ignore previous instructions'. These are used to clear the agent's context during multi-phase workflows and specifically to prevent 'self-injection' from earlier outputs, according to the 'Generator-Evaluator Contract Protocol'.
[EXTERNAL_DOWNLOADS]: The README and installation guides provide the official setup command for the platform (curl -fsSL https://claude.ai/install.sh | bash), which is a standard procedure for installing well-known developer tooling.

academic-research-suite