The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and synthesizes information from external research sources and user-provided documents. This is a standard risk for research agents, which is well-mitigated here through explicit workflow instructions:
Ingestion points: The agent processes user-supplied research questions, external web search results, and provided reference papers (SKILL.md, references/report.md).
Boundary markers: The skill mandates the use of 'Facts/Assumptions/Decisions/Risks' (FADR) separation and confidence-tagging (H/M/L) to maintain clear boundaries between external evidence and model synthesis.
Capability inventory: Capabilities include file system writes for report generation (docs/research/) and the use of sub-agents for parallel research fan-out.
Sanitization: A dedicated 'adversarial pass' (Step 5 in SKILL.md) requires the agent to re-verify every cited source and drop any unsupported claims, specifically targeting potential hallucinations or data-poisoning in ingested sources.
[COMMAND_EXECUTION]: The evals/run-static-checks.sh script executes shell commands to trigger a Python-based static analysis tool. This is a development-time utility provided to verify the integrity of the skill's reference files and metadata; it is not part of the agent's runtime behavior during research tasks.
[EXTERNAL_DOWNLOADS]: The skill's primary function is to search for and retrieve information from external sources (industry reports, academic databases, etc.). These operations are consistent with the skill's stated purpose and rely on standard agent tools for information retrieval.

research