The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to perform web searches and fetch content from live university websites, program calls, and pathway descriptions to perform 'Thematic Fit' assessments. This process ingests untrusted data that could potentially contain malicious instructions intended to mislead the agent or compromise user-provided data. \n- Ingestion points: As described in the Workflow (Step 2) of SKILL.md and the Extraction Method in references/application-regimes.md, the agent is required to 'Search the web for every target' and 'Fetch the programme/consortium page, the current cycle's call, and each pathway page separately'. \n- Boundary markers: The instructions do not mandate the use of explicit delimiters (e.g., XML tags or triple quotes) to isolate external content, nor do they provide a system-level warning to ignore embedded instructions in the fetched text. \n- Capability inventory: The skill leverages web search and fetching tools to gather data, which is then processed into fit classifications, bridging sentences, and SOP notes. \n- Sanitization: While the skill provides logical 'guards' such as the 'swap test' and 'facts fixed' rule to ensure assessments are evidence-based, it lacks technical sanitization or filtering to prevent the execution of instructions embedded in the external content.

apply-program-fit-mapper