The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze user-provided requirement statements. This presents a potential surface for indirect prompt injection if the input contains instructions designed to manipulate the agent's output.
Ingestion points: User-provided requirement wording processed via the instructions in prompts/requirement-ambiguity-analysis.md.
Boundary markers: The instructions emphasize using "only supplied material" and strictly adhering to an analyst role with defined output schemas.
Capability inventory: The skill performs text-based analysis; no dangerous file-system or network capabilities are exposed to the prompt.
Sanitization: No explicit input sanitization is present; however, the agent is instructed to diagnose under-specification rather than execute instructions.
[COMMAND_EXECUTION]: The repository includes local scripts used for regression testing of the analysis logic.
Evidence: evals/fixtures/check-domain-decisions.sh and evals/fixtures/check_domain_decisions.py are used to verify the correctness of markdown tables generated during evaluation.
The Python script performs deterministic checks on output structure and content using standard libraries (re, json, decimal) and does not use unsafe execution functions like eval() or exec() on external data.

requirement-ambiguity-analysis